xkrs_manan/RODY
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

first commit

Browse Source
This commit is contained in:
552068321@qq.com
2022-11-04 17:37:08 +08:00
commit 6f7de660aa
192 changed files with 32574 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

65
app/yolov5/utils/__init__.py Normal file
View File

@ -0,0 +1,65 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
utils/initialization
"""
import contextlib
import threading
class TryExcept(contextlib.ContextDecorator):
# YOLOv5 TryExcept class. Usage: @TryExcept() decorator or 'with TryExcept():' context manager
def __init__(self, msg=''):
self.msg = msg
def __enter__(self):
pass
def __exit__(self, exc_type, value, traceback):
if value:
print(f'{self.msg}{value}')
return True
def threaded(func):
# Multi-threads a target function and returns thread. Usage: @threaded decorator
def wrapper(*args, **kwargs):
thread = threading.Thread(target=func, args=args, kwargs=kwargs, daemon=True)
thread.start()
return thread
return wrapper
def notebook_init(verbose=True):
# Check system software and hardware
print('Checking setup...')
import os
import shutil
from utils.general import check_font, check_requirements, emojis, is_colab
from utils.torch_utils import select_device # imports
check_requirements(('psutil', 'IPython'))
check_font()
import psutil
from IPython import display # to display images and clear console output
if is_colab():
shutil.rmtree('/content/sample_data', ignore_errors=True) # remove colab /sample_data directory
# System info
if verbose:
gb = 1 << 30 # bytes to GiB (1024 ** 3)
ram = psutil.virtual_memory().total
total, used, free = shutil.disk_usage("/")
display.clear_output()
s = f'({os.cpu_count()} CPUs, {ram / gb:.1f} GB RAM, {(total - free) / gb:.1f}/{total / gb:.1f} GB disk)'
else:
s = ''
select_device(newline=False)
print(emojis(f'Setup complete ✅ {s}'))
return display

103
app/yolov5/utils/activations.py Normal file
View File

@ -0,0 +1,103 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Activation functions
"""
import torch
import torch.nn as nn
import torch.nn.functional as F
class SiLU(nn.Module):
# SiLU activation https://arxiv.org/pdf/1606.08415.pdf
@staticmethod
def forward(x):
return x * torch.sigmoid(x)
class Hardswish(nn.Module):
# Hard-SiLU activation
@staticmethod
def forward(x):
# return x * F.hardsigmoid(x) # for TorchScript and CoreML
return x * F.hardtanh(x + 3, 0.0, 6.0) / 6.0 # for TorchScript, CoreML and ONNX
class Mish(nn.Module):
# Mish activation https://github.com/digantamisra98/Mish
@staticmethod
def forward(x):
return x * F.softplus(x).tanh()
class MemoryEfficientMish(nn.Module):
# Mish activation memory-efficient
class F(torch.autograd.Function):
@staticmethod
def forward(ctx, x):
ctx.save_for_backward(x)
return x.mul(torch.tanh(F.softplus(x))) # x * tanh(ln(1 + exp(x)))
@staticmethod
def backward(ctx, grad_output):
x = ctx.saved_tensors[0]
sx = torch.sigmoid(x)
fx = F.softplus(x).tanh()
return grad_output * (fx + x * sx * (1 - fx * fx))
def forward(self, x):
return self.F.apply(x)
class FReLU(nn.Module):
# FReLU activation https://arxiv.org/abs/2007.11824
def __init__(self, c1, k=3): # ch_in, kernel
super().__init__()
self.conv = nn.Conv2d(c1, c1, k, 1, 1, groups=c1, bias=False)
self.bn = nn.BatchNorm2d(c1)
def forward(self, x):
return torch.max(x, self.bn(self.conv(x)))
class AconC(nn.Module):
r""" ACON activation (activate or not)
AconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is a learnable parameter
according to "Activate or Not: Learning Customized Activation" <https://arxiv.org/pdf/2009.04759.pdf>.
"""
def __init__(self, c1):
super().__init__()
self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1))
self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1))
self.beta = nn.Parameter(torch.ones(1, c1, 1, 1))
def forward(self, x):
dpx = (self.p1 - self.p2) * x
return dpx * torch.sigmoid(self.beta * dpx) + self.p2 * x
class MetaAconC(nn.Module):
r""" ACON activation (activate or not)
MetaAconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is generated by a small network
according to "Activate or Not: Learning Customized Activation" <https://arxiv.org/pdf/2009.04759.pdf>.
"""
def __init__(self, c1, k=1, s=1, r=16): # ch_in, kernel, stride, r
super().__init__()
c2 = max(r, c1 // r)
self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1))
self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1))
self.fc1 = nn.Conv2d(c1, c2, k, s, bias=True)
self.fc2 = nn.Conv2d(c2, c1, k, s, bias=True)
# self.bn1 = nn.BatchNorm2d(c2)
# self.bn2 = nn.BatchNorm2d(c1)
def forward(self, x):
y = x.mean(dim=2, keepdims=True).mean(dim=3, keepdims=True)
# batch-size 1 bug/instabilities https://github.com/ultralytics/yolov5/issues/2891
# beta = torch.sigmoid(self.bn2(self.fc2(self.bn1(self.fc1(y))))) # bug/unstable
beta = torch.sigmoid(self.fc2(self.fc1(y))) # bug patch BN layers removed
dpx = (self.p1 - self.p2) * x
return dpx * torch.sigmoid(beta * dpx) + self.p2 * x

396
app/yolov5/utils/augmentations.py Normal file
View File

@ -0,0 +1,396 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Image augmentation functions
"""
import math
import random
import cv2
import numpy as np
import torch
import torchvision.transforms as T
import torchvision.transforms.functional as TF
from app.yolov5.utils.general import LOGGER, check_version, colorstr, resample_segments, segment2box
from app.yolov5.utils.metrics import bbox_ioa
IMAGENET_MEAN = 0.485, 0.456, 0.406 # RGB mean
IMAGENET_STD = 0.229, 0.224, 0.225 # RGB standard deviation
class Albumentations:
# YOLOv5 Albumentations class (optional, only used if package is installed)
def __init__(self):
self.transform = None
prefix = colorstr('albumentations: ')
try:
import albumentations as A
check_version(A.__version__, '1.0.3', hard=True) # version requirement
T = [
A.Blur(p=0.01),
A.MedianBlur(p=0.01),
A.ToGray(p=0.01),
A.CLAHE(p=0.01),
A.RandomBrightnessContrast(p=0.0),
A.RandomGamma(p=0.0),
A.ImageCompression(quality_lower=75, p=0.0)] # transforms
self.transform = A.Compose(T, bbox_params=A.BboxParams(format='yolo', label_fields=['class_labels']))
LOGGER.info(prefix + ', '.join(f'{x}'.replace('always_apply=False, ', '') for x in T if x.p))
except ImportError: # package not installed, skip
pass
except Exception as e:
LOGGER.info(f'{prefix}{e}')
def __call__(self, im, labels, p=1.0):
if self.transform and random.random() < p:
new = self.transform(image=im, bboxes=labels[:, 1:], class_labels=labels[:, 0]) # transformed
im, labels = new['image'], np.array([[c, *b] for c, b in zip(new['class_labels'], new['bboxes'])])
return im, labels
def normalize(x, mean=IMAGENET_MEAN, std=IMAGENET_STD, inplace=False):
# Denormalize RGB images x per ImageNet stats in BCHW format, i.e. = (x - mean) / std
return TF.normalize(x, mean, std, inplace=inplace)
def denormalize(x, mean=IMAGENET_MEAN, std=IMAGENET_STD):
# Denormalize RGB images x per ImageNet stats in BCHW format, i.e. = x * std + mean
for i in range(3):
x[:, i] = x[:, i] * std[i] + mean[i]
return x
def augment_hsv(im, hgain=0.5, sgain=0.5, vgain=0.5):
# HSV color-space augmentation
if hgain or sgain or vgain:
r = np.random.uniform(-1, 1, 3) * [hgain, sgain, vgain] + 1 # random gains
hue, sat, val = cv2.split(cv2.cvtColor(im, cv2.COLOR_BGR2HSV))
dtype = im.dtype # uint8
x = np.arange(0, 256, dtype=r.dtype)
lut_hue = ((x * r[0]) % 180).astype(dtype)
lut_sat = np.clip(x * r[1], 0, 255).astype(dtype)
lut_val = np.clip(x * r[2], 0, 255).astype(dtype)
im_hsv = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val)))
cv2.cvtColor(im_hsv, cv2.COLOR_HSV2BGR, dst=im) # no return needed
def hist_equalize(im, clahe=True, bgr=False):
# Equalize histogram on BGR image 'im' with im.shape(n,m,3) and range 0-255
yuv = cv2.cvtColor(im, cv2.COLOR_BGR2YUV if bgr else cv2.COLOR_RGB2YUV)
if clahe:
c = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
yuv[:, :, 0] = c.apply(yuv[:, :, 0])
else:
yuv[:, :, 0] = cv2.equalizeHist(yuv[:, :, 0]) # equalize Y channel histogram
return cv2.cvtColor(yuv, cv2.COLOR_YUV2BGR if bgr else cv2.COLOR_YUV2RGB) # convert YUV image to RGB
def replicate(im, labels):
# Replicate labels
h, w = im.shape[:2]
boxes = labels[:, 1:].astype(int)
x1, y1, x2, y2 = boxes.T
s = ((x2 - x1) + (y2 - y1)) / 2 # side length (pixels)
for i in s.argsort()[:round(s.size * 0.5)]: # smallest indices
x1b, y1b, x2b, y2b = boxes[i]
bh, bw = y2b - y1b, x2b - x1b
yc, xc = int(random.uniform(0, h - bh)), int(random.uniform(0, w - bw)) # offset x, y
x1a, y1a, x2a, y2a = [xc, yc, xc + bw, yc + bh]
im[y1a:y2a, x1a:x2a] = im[y1b:y2b, x1b:x2b] # im4[ymin:ymax, xmin:xmax]
labels = np.append(labels, [[labels[i, 0], x1a, y1a, x2a, y2a]], axis=0)
return im, labels
def letterbox(im, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32):
# Resize and pad image while meeting stride-multiple constraints
shape = im.shape[:2] # current shape [height, width]
if isinstance(new_shape, int):
new_shape = (new_shape, new_shape)
# Scale ratio (new / old)
r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
if not scaleup: # only scale down, do not scale up (for better val mAP)
r = min(r, 1.0)
# Compute padding
ratio = r, r # width, height ratios
new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))
dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1] # wh padding
if auto: # minimum rectangle
dw, dh = np.mod(dw, stride), np.mod(dh, stride) # wh padding
elif scaleFill: # stretch
dw, dh = 0.0, 0.0
new_unpad = (new_shape[1], new_shape[0])
ratio = new_shape[1] / shape[1], new_shape[0] / shape[0] # width, height ratios
dw /= 2 # divide padding into 2 sides
dh /= 2
if shape[::-1] != new_unpad: # resize
im = cv2.resize(im, new_unpad, interpolation=cv2.INTER_LINEAR)
top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
im = cv2.copyMakeBorder(im, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color) # add border
return im, ratio, (dw, dh)
def random_perspective(im,
targets=(),
segments=(),
degrees=10,
translate=.1,
scale=.1,
shear=10,
perspective=0.0,
border=(0, 0)):
# torchvision.transforms.RandomAffine(degrees=(-10, 10), translate=(0.1, 0.1), scale=(0.9, 1.1), shear=(-10, 10))
# targets = [cls, xyxy]
height = im.shape[0] + border[0] * 2 # shape(h,w,c)
width = im.shape[1] + border[1] * 2
# Center
C = np.eye(3)
C[0, 2] = -im.shape[1] / 2 # x translation (pixels)
C[1, 2] = -im.shape[0] / 2 # y translation (pixels)
# Perspective
P = np.eye(3)
P[2, 0] = random.uniform(-perspective, perspective) # x perspective (about y)
P[2, 1] = random.uniform(-perspective, perspective) # y perspective (about x)
# Rotation and Scale
R = np.eye(3)
a = random.uniform(-degrees, degrees)
# a += random.choice([-180, -90, 0, 90]) # add 90deg rotations to small rotations
s = random.uniform(1 - scale, 1 + scale)
# s = 2 ** random.uniform(-scale, scale)
R[:2] = cv2.getRotationMatrix2D(angle=a, center=(0, 0), scale=s)
# Shear
S = np.eye(3)
S[0, 1] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # x shear (deg)
S[1, 0] = math.tan(random.uniform(-shear, shear) * math.pi / 180) # y shear (deg)
# Translation
T = np.eye(3)
T[0, 2] = random.uniform(0.5 - translate, 0.5 + translate) * width # x translation (pixels)
T[1, 2] = random.uniform(0.5 - translate, 0.5 + translate) * height # y translation (pixels)
# Combined rotation matrix
M = T @ S @ R @ P @ C # order of operations (right to left) is IMPORTANT
if (border[0] != 0) or (border[1] != 0) or (M != np.eye(3)).any(): # image changed
if perspective:
im = cv2.warpPerspective(im, M, dsize=(width, height), borderValue=(114, 114, 114))
else: # affine
im = cv2.warpAffine(im, M[:2], dsize=(width, height), borderValue=(114, 114, 114))
# Visualize
# import matplotlib.pyplot as plt
# ax = plt.subplots(1, 2, figsize=(12, 6))[1].ravel()
# ax[0].imshow(im[:, :, ::-1]) # base
# ax[1].imshow(im2[:, :, ::-1]) # warped
# Transform label coordinates
n = len(targets)
if n:
use_segments = any(x.any() for x in segments)
new = np.zeros((n, 4))
if use_segments: # warp segments
segments = resample_segments(segments) # upsample
for i, segment in enumerate(segments):
xy = np.ones((len(segment), 3))
xy[:, :2] = segment
xy = xy @ M.T # transform
xy = xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2] # perspective rescale or affine
# clip
new[i] = segment2box(xy, width, height)
else: # warp boxes
xy = np.ones((n * 4, 3))
xy[:, :2] = targets[:, [1, 2, 3, 4, 1, 4, 3, 2]].reshape(n * 4, 2) # x1y1, x2y2, x1y2, x2y1
xy = xy @ M.T # transform
xy = (xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2]).reshape(n, 8) # perspective rescale or affine
# create new boxes
x = xy[:, [0, 2, 4, 6]]
y = xy[:, [1, 3, 5, 7]]
new = np.concatenate((x.min(1), y.min(1), x.max(1), y.max(1))).reshape(4, n).T
# clip
new[:, [0, 2]] = new[:, [0, 2]].clip(0, width)
new[:, [1, 3]] = new[:, [1, 3]].clip(0, height)
# filter candidates
i = box_candidates(box1=targets[:, 1:5].T * s, box2=new.T, area_thr=0.01 if use_segments else 0.10)
targets = targets[i]
targets[:, 1:5] = new[i]
return im, targets
def copy_paste(im, labels, segments, p=0.5):
# Implement Copy-Paste augmentation https://arxiv.org/abs/2012.07177, labels as nx5 np.array(cls, xyxy)
n = len(segments)
if p and n:
h, w, c = im.shape # height, width, channels
im_new = np.zeros(im.shape, np.uint8)
for j in random.sample(range(n), k=round(p * n)):
l, s = labels[j], segments[j]
box = w - l[3], l[2], w - l[1], l[4]
ioa = bbox_ioa(box, labels[:, 1:5]) # intersection over area
if (ioa < 0.30).all(): # allow 30% obscuration of existing labels
labels = np.concatenate((labels, [[l[0], *box]]), 0)
segments.append(np.concatenate((w - s[:, 0:1], s[:, 1:2]), 1))
cv2.drawContours(im_new, [segments[j].astype(np.int32)], -1, (255, 255, 255), cv2.FILLED)
result = cv2.bitwise_and(src1=im, src2=im_new)
result = cv2.flip(result, 1) # augment segments (flip left-right)
i = result > 0 # pixels to replace
# i[:, :] = result.max(2).reshape(h, w, 1) # act over ch
im[i] = result[i] # cv2.imwrite('debug.jpg', im) # debug
return im, labels, segments
def cutout(im, labels, p=0.5):
# Applies image cutout augmentation https://arxiv.org/abs/1708.04552
if random.random() < p:
h, w = im.shape[:2]
scales = [0.5] * 1 + [0.25] * 2 + [0.125] * 4 + [0.0625] * 8 + [0.03125] * 16 # image size fraction
for s in scales:
mask_h = random.randint(1, int(h * s)) # create random masks
mask_w = random.randint(1, int(w * s))
# box
xmin = max(0, random.randint(0, w) - mask_w // 2)
ymin = max(0, random.randint(0, h) - mask_h // 2)
xmax = min(w, xmin + mask_w)
ymax = min(h, ymin + mask_h)
# apply random color mask
im[ymin:ymax, xmin:xmax] = [random.randint(64, 191) for _ in range(3)]
# return unobscured labels
if len(labels) and s > 0.03:
box = np.array([xmin, ymin, xmax, ymax], dtype=np.float32)
ioa = bbox_ioa(box, labels[:, 1:5]) # intersection over area
labels = labels[ioa < 0.60] # remove >60% obscured labels
return labels
def mixup(im, labels, im2, labels2):
# Applies MixUp augmentation https://arxiv.org/pdf/1710.09412.pdf
r = np.random.beta(32.0, 32.0) # mixup ratio, alpha=beta=32.0
im = (im * r + im2 * (1 - r)).astype(np.uint8)
labels = np.concatenate((labels, labels2), 0)
return im, labels
def box_candidates(box1, box2, wh_thr=2, ar_thr=100, area_thr=0.1, eps=1e-16): # box1(4,n), box2(4,n)
# Compute candidate boxes: box1 before augment, box2 after augment, wh_thr (pixels), aspect_ratio_thr, area_ratio
w1, h1 = box1[2] - box1[0], box1[3] - box1[1]
w2, h2 = box2[2] - box2[0], box2[3] - box2[1]
ar = np.maximum(w2 / (h2 + eps), h2 / (w2 + eps)) # aspect ratio
return (w2 > wh_thr) & (h2 > wh_thr) & (w2 * h2 / (w1 * h1 + eps) > area_thr) & (ar < ar_thr) # candidates
def classify_albumentations(augment=True,
size=224,
scale=(0.08, 1.0),
hflip=0.5,
vflip=0.0,
jitter=0.4,
mean=IMAGENET_MEAN,
std=IMAGENET_STD,
auto_aug=False):
# YOLOv5 classification Albumentations (optional, only used if package is installed)
prefix = colorstr('albumentations: ')
try:
import albumentations as A
from albumentations.pytorch import ToTensorV2
check_version(A.__version__, '1.0.3', hard=True) # version requirement
if augment: # Resize and crop
T = [A.RandomResizedCrop(height=size, width=size, scale=scale)]
if auto_aug:
# TODO: implement AugMix, AutoAug & RandAug in albumentation
LOGGER.info(f'{prefix}auto augmentations are currently not supported')
else:
if hflip > 0:
T += [A.HorizontalFlip(p=hflip)]
if vflip > 0:
T += [A.VerticalFlip(p=vflip)]
if jitter > 0:
color_jitter = (float(jitter),) * 3 # repeat value for brightness, contrast, satuaration, 0 hue
T += [A.ColorJitter(*color_jitter, 0)]
else: # Use fixed crop for eval set (reproducibility)
T = [A.SmallestMaxSize(max_size=size), A.CenterCrop(height=size, width=size)]
T += [A.Normalize(mean=mean, std=std), ToTensorV2()] # Normalize and convert to Tensor
LOGGER.info(prefix + ', '.join(f'{x}'.replace('always_apply=False, ', '') for x in T if x.p))
return A.Compose(T)
except ImportError: # package not installed, skip
pass
except Exception as e:
LOGGER.info(f'{prefix}{e}')
def classify_transforms(size=224):
# Transforms to apply if albumentations not installed
assert isinstance(size, int), f'ERROR: classify_transforms size {size} must be integer, not (list, tuple)'
# T.Compose([T.ToTensor(), T.Resize(size), T.CenterCrop(size), T.Normalize(IMAGENET_MEAN, IMAGENET_STD)])
return T.Compose([CenterCrop(size), ToTensor(), T.Normalize(IMAGENET_MEAN, IMAGENET_STD)])
class LetterBox:
# YOLOv5 LetterBox class for image preprocessing, i.e. T.Compose([LetterBox(size), ToTensor()])
def __init__(self, size=(640, 640), auto=False, stride=32):
super().__init__()
self.h, self.w = (size, size) if isinstance(size, int) else size
self.auto = auto # pass max size integer, automatically solve for short side using stride
self.stride = stride # used with auto
def __call__(self, im): # im = np.array HWC
imh, imw = im.shape[:2]
r = min(self.h / imh, self.w / imw) # ratio of new/old
h, w = round(imh * r), round(imw * r) # resized image
hs, ws = (math.ceil(x / self.stride) * self.stride for x in (h, w)) if self.auto else self.h, self.w
top, left = round((hs - h) / 2 - 0.1), round((ws - w) / 2 - 0.1)
im_out = np.full((self.h, self.w, 3), 114, dtype=im.dtype)
im_out[top:top + h, left:left + w] = cv2.resize(im, (w, h), interpolation=cv2.INTER_LINEAR)
return im_out
class CenterCrop:
# YOLOv5 CenterCrop class for image preprocessing, i.e. T.Compose([CenterCrop(size), ToTensor()])
def __init__(self, size=640):
super().__init__()
self.h, self.w = (size, size) if isinstance(size, int) else size
def __call__(self, im): # im = np.array HWC
imh, imw = im.shape[:2]
m = min(imh, imw) # min dimension
top, left = (imh - m) // 2, (imw - m) // 2
return cv2.resize(im[top:top + m, left:left + m], (self.w, self.h), interpolation=cv2.INTER_LINEAR)
class ToTensor:
# YOLOv5 ToTensor class for image preprocessing, i.e. T.Compose([LetterBox(size), ToTensor()])
def __init__(self, half=False):
super().__init__()
self.half = half
def __call__(self, im): # im = np.array HWC in BGR order
im = np.ascontiguousarray(im.transpose((2, 0, 1))[::-1]) # HWC to CHW -> BGR to RGB -> contiguous
im = torch.from_numpy(im) # to torch
im = im.half() if self.half else im.float() # uint8 to fp16/32
im /= 255.0 # 0-255 to 0.0-1.0
return im

169
app/yolov5/utils/autoanchor.py Normal file
View File

@ -0,0 +1,169 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
AutoAnchor utils
"""
import random
import numpy as np
import torch
import yaml
from tqdm import tqdm
from app.yolov5.utils import TryExcept
from app.yolov5.utils.general import LOGGER, colorstr
PREFIX = colorstr('AutoAnchor: ')
def check_anchor_order(m):
# Check anchor order against stride order for YOLOv5 Detect() module m, and correct if necessary
a = m.anchors.prod(-1).mean(-1).view(-1) # mean anchor area per output layer
da = a[-1] - a[0] # delta a
ds = m.stride[-1] - m.stride[0] # delta s
if da and (da.sign() != ds.sign()): # same order
LOGGER.info(f'{PREFIX}Reversing anchor order')
m.anchors[:] = m.anchors.flip(0)
@TryExcept(f'{PREFIX}ERROR: ')
def check_anchors(dataset, model, thr=4.0, imgsz=640):
# Check anchor fit to data, recompute if necessary
m = model.module.model[-1] if hasattr(model, 'module') else model.model[-1] # Detect()
shapes = imgsz * dataset.shapes / dataset.shapes.max(1, keepdims=True)
scale = np.random.uniform(0.9, 1.1, size=(shapes.shape[0], 1)) # augment scale
wh = torch.tensor(np.concatenate([l[:, 3:5] * s for s, l in zip(shapes * scale, dataset.labels)])).float() # wh
def metric(k): # compute metric
r = wh[:, None] / k[None]
x = torch.min(r, 1 / r).min(2)[0] # ratio metric
best = x.max(1)[0] # best_x
aat = (x > 1 / thr).float().sum(1).mean() # anchors above threshold
bpr = (best > 1 / thr).float().mean() # best possible recall
return bpr, aat
stride = m.stride.to(m.anchors.device).view(-1, 1, 1) # model strides
anchors = m.anchors.clone() * stride # current anchors
bpr, aat = metric(anchors.cpu().view(-1, 2))
s = f'\n{PREFIX}{aat:.2f} anchors/target, {bpr:.3f} Best Possible Recall (BPR). '
if bpr > 0.98: # threshold to recompute
LOGGER.info(f'{s}Current anchors are a good fit to dataset ✅')
else:
LOGGER.info(f'{s}Anchors are a poor fit to dataset ⚠️, attempting to improve...')
na = m.anchors.numel() // 2 # number of anchors
anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False)
new_bpr = metric(anchors)[0]
if new_bpr > bpr: # replace anchors
anchors = torch.tensor(anchors, device=m.anchors.device).type_as(m.anchors)
m.anchors[:] = anchors.clone().view_as(m.anchors)
check_anchor_order(m) # must be in pixel-space (not grid-space)
m.anchors /= stride
s = f'{PREFIX}Done ✅ (optional: update model *.yaml to use these anchors in the future)'
else:
s = f'{PREFIX}Done ⚠️ (original anchors better than new anchors, proceeding with original anchors)'
LOGGER.info(s)
def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=1000, verbose=True):
""" Creates kmeans-evolved anchors from training dataset
Arguments:
dataset: path to data.yaml, or a loaded dataset
n: number of anchors
img_size: image size used for training
thr: anchor-label wh ratio threshold hyperparameter hyp['anchor_t'] used for training, default=4.0
gen: generations to evolve anchors using genetic algorithm
verbose: print all results
Return:
k: kmeans evolved anchors
Usage:
from utils.autoanchor import *; _ = kmean_anchors()
"""
from scipy.cluster.vq import kmeans
npr = np.random
thr = 1 / thr
def metric(k, wh): # compute metrics
r = wh[:, None] / k[None]
x = torch.min(r, 1 / r).min(2)[0] # ratio metric
# x = wh_iou(wh, torch.tensor(k)) # iou metric
return x, x.max(1)[0] # x, best_x
def anchor_fitness(k): # mutation fitness
_, best = metric(torch.tensor(k, dtype=torch.float32), wh)
return (best * (best > thr).float()).mean() # fitness
def print_results(k, verbose=True):
k = k[np.argsort(k.prod(1))] # sort small to large
x, best = metric(k, wh0)
bpr, aat = (best > thr).float().mean(), (x > thr).float().mean() * n # best possible recall, anch > thr
s = f'{PREFIX}thr={thr:.2f}: {bpr:.4f} best possible recall, {aat:.2f} anchors past thr\n' \
f'{PREFIX}n={n}, img_size={img_size}, metric_all={x.mean():.3f}/{best.mean():.3f}-mean/best, ' \
f'past_thr={x[x > thr].mean():.3f}-mean: '
for x in k:
s += '%i,%i, ' % (round(x[0]), round(x[1]))
if verbose:
LOGGER.info(s[:-2])
return k
if isinstance(dataset, str): # *.yaml file
with open(dataset, errors='ignore') as f:
data_dict = yaml.safe_load(f) # model dict
from utils.dataloaders import LoadImagesAndLabels
dataset = LoadImagesAndLabels(data_dict['train'], augment=True, rect=True)
# Get label wh
shapes = img_size * dataset.shapes / dataset.shapes.max(1, keepdims=True)
wh0 = np.concatenate([l[:, 3:5] * s for s, l in zip(shapes, dataset.labels)]) # wh
# Filter
i = (wh0 < 3.0).any(1).sum()
if i:
LOGGER.info(f'{PREFIX}WARNING: Extremely small objects found: {i} of {len(wh0)} labels are < 3 pixels in size')
wh = wh0[(wh0 >= 2.0).any(1)].astype(np.float32) # filter > 2 pixels
# wh = wh * (npr.rand(wh.shape[0], 1) * 0.9 + 0.1) # multiply by random scale 0-1
# Kmeans init
try:
LOGGER.info(f'{PREFIX}Running kmeans for {n} anchors on {len(wh)} points...')
assert n <= len(wh) # apply overdetermined constraint
s = wh.std(0) # sigmas for whitening
k = kmeans(wh / s, n, iter=30)[0] * s # points
assert n == len(k) # kmeans may return fewer points than requested if wh is insufficient or too similar
except Exception:
LOGGER.warning(f'{PREFIX}WARNING: switching strategies from kmeans to random init')
k = np.sort(npr.rand(n * 2)).reshape(n, 2) * img_size # random init
wh, wh0 = (torch.tensor(x, dtype=torch.float32) for x in (wh, wh0))
k = print_results(k, verbose=False)
# Plot
# k, d = [None] * 20, [None] * 20
# for i in tqdm(range(1, 21)):
# k[i-1], d[i-1] = kmeans(wh / s, i) # points, mean distance
# fig, ax = plt.subplots(1, 2, figsize=(14, 7), tight_layout=True)
# ax = ax.ravel()
# ax[0].plot(np.arange(1, 21), np.array(d) ** 2, marker='.')
# fig, ax = plt.subplots(1, 2, figsize=(14, 7)) # plot wh
# ax[0].hist(wh[wh[:, 0]<100, 0],400)
# ax[1].hist(wh[wh[:, 1]<100, 1],400)
# fig.savefig('wh.png', dpi=200)
# Evolve
f, sh, mp, s = anchor_fitness(k), k.shape, 0.9, 0.1 # fitness, generations, mutation prob, sigma
pbar = tqdm(range(gen), bar_format='{l_bar}{bar:10}{r_bar}{bar:-10b}') # progress bar
for _ in pbar:
v = np.ones(sh)
while (v == 1).all(): # mutate until a change occurs (prevent duplicates)
v = ((npr.random(sh) < mp) * random.random() * npr.randn(*sh) * s + 1).clip(0.3, 3.0)
kg = (k.copy() * v).clip(min=2.0)
fg = anchor_fitness(kg)
if fg > f:
f, k = fg, kg.copy()
pbar.desc = f'{PREFIX}Evolving anchors with Genetic Algorithm: fitness = {f:.4f}'
if verbose:
print_results(k, verbose)
return print_results(k).astype(np.float32)

69
app/yolov5/utils/autobatch.py Normal file
View File

@ -0,0 +1,69 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Auto-batch utils
"""
from copy import deepcopy
import numpy as np
import torch
from app.yolov5.utils.general import LOGGER, colorstr
from app.yolov5.utils.torch_utils import profile
def check_train_batch_size(model, imgsz=640, amp=True):
# Check YOLOv5 training batch size
with torch.cuda.amp.autocast(amp):
return autobatch(deepcopy(model).train(), imgsz) # compute optimal batch size
def autobatch(model, imgsz=640, fraction=0.8, batch_size=16):
# Automatically estimate best batch size to use `fraction` of available CUDA memory
# Usage:
# import torch
# from utils.autobatch import autobatch
# model = torch.hub.load('ultralytics/yolov5', 'yolov5s', autoshape=False)
# print(autobatch(model))
# Check device
prefix = colorstr('AutoBatch: ')
LOGGER.info(f'{prefix}Computing optimal batch size for --imgsz {imgsz}')
device = next(model.parameters()).device # get model device
if device.type == 'cpu':
LOGGER.info(f'{prefix}CUDA not detected, using default CPU batch-size {batch_size}')
return batch_size
# Inspect CUDA memory
gb = 1 << 30 # bytes to GiB (1024 ** 3)
d = str(device).upper() # 'CUDA:0'
properties = torch.cuda.get_device_properties(device) # device properties
t = properties.total_memory / gb # GiB total
r = torch.cuda.memory_reserved(device) / gb # GiB reserved
a = torch.cuda.memory_allocated(device) / gb # GiB allocated
f = t - (r + a) # GiB free
LOGGER.info(f'{prefix}{d} ({properties.name}) {t:.2f}G total, {r:.2f}G reserved, {a:.2f}G allocated, {f:.2f}G free')
# Profile batch sizes
batch_sizes = [1, 2, 4, 8, 16]
try:
img = [torch.empty(b, 3, imgsz, imgsz) for b in batch_sizes]
results = profile(img, model, n=3, device=device)
except Exception as e:
LOGGER.warning(f'{prefix}{e}')
# Fit a solution
y = [x[2] for x in results if x] # memory [2]
p = np.polyfit(batch_sizes[:len(y)], y, deg=1) # first degree polynomial fit
b = int((f * fraction - p[1]) / p[0]) # y intercept (optimal batch size)
if None in results: # some sizes failed
i = results.index(None) # first fail index
if b >= batch_sizes[i]: # y intercept above failure point
b = batch_sizes[max(i - 1, 0)] # select prior safe point
if b < 1 or b > 1024: # b outside of safe range
b = batch_size
LOGGER.warning(f'{prefix}WARNING: ⚠️ CUDA anomaly detected, recommend restart environment and retry command.')
fraction = np.polyval(p, b) / t # actual fraction predicted
LOGGER.info(f'{prefix}Using batch-size {b} for {d} {t * fraction:.2f}G/{t:.2f}G ({fraction * 100:.0f}%) ✅')
return b

0
app/yolov5/utils/aws/__init__.py Normal file
View File

26
app/yolov5/utils/aws/mime.sh Normal file
View File

@ -0,0 +1,26 @@
# AWS EC2 instance startup 'MIME' script https://aws.amazon.com/premiumsupport/knowledge-center/execute-user-data-ec2/
# This script will run on every instance restart, not only on first start
# --- DO NOT COPY ABOVE COMMENTS WHEN PASTING INTO USERDATA ---
Content-Type: multipart/mixed; boundary="//"
MIME-Version: 1.0
--//
Content-Type: text/cloud-config; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment; filename="cloud-config.txt"
#cloud-config
cloud_final_modules:
- [scripts-user, always]
--//
Content-Type: text/x-shellscript; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment; filename="userdata.txt"
#!/bin/bash
# --- paste contents of userdata.sh here ---
--//

40
app/yolov5/utils/aws/resume.py Normal file
View File

@ -0,0 +1,40 @@
# Resume all interrupted trainings in yolov5/ dir including DDP trainings
# Usage: $ python utils/aws/resume.py
import os
import sys
from pathlib import Path
import torch
import yaml
FILE = Path(__file__).resolve()
ROOT = FILE.parents[2] # YOLOv5 root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
port = 0 # --master_port
path = Path('').resolve()
for last in path.rglob('*/**/last.pt'):
ckpt = torch.load(last)
if ckpt['optimizer'] is None:
continue
# Load opt.yaml
with open(last.parent.parent / 'opt.yaml', errors='ignore') as f:
opt = yaml.safe_load(f)
# Get device count
d = opt['device'].split(',') # devices
nd = len(d) # number of devices
ddp = nd > 1 or (nd == 0 and torch.cuda.device_count() > 1) # distributed data parallel
if ddp: # multi-GPU
port += 1
cmd = f'python -m torch.distributed.run --nproc_per_node {nd} --master_port {port} train.py --resume {last}'
else: # single-GPU
cmd = f'python train.py --resume {last}'
cmd += ' > /dev/null 2>&1 &' # redirect output to dev/null and run in daemon thread
print(cmd)
os.system(cmd)

27
app/yolov5/utils/aws/userdata.sh Normal file
View File

@ -0,0 +1,27 @@
#!/bin/bash
# AWS EC2 instance startup script https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/user-data.html
# This script will run only once on first instance start (for a re-start script see mime.sh)
# /home/ubuntu (ubuntu) or /home/ec2-user (amazon-linux) is working dir
# Use >300 GB SSD
cd home/ubuntu
if [ ! -d yolov5 ]; then
echo "Running first-time script." # install dependencies, download COCO, pull Docker
git clone https://github.com/ultralytics/yolov5 -b master && sudo chmod -R 777 yolov5
cd yolov5
bash data/scripts/get_coco.sh && echo "COCO done." &
sudo docker pull ultralytics/yolov5:latest && echo "Docker done." &
python -m pip install --upgrade pip && pip install -r requirements.txt && python detect.py && echo "Requirements done." &
wait && echo "All tasks done." # finish background tasks
else
echo "Running re-start script." # resume interrupted runs
i=0
list=$(sudo docker ps -qa) # container list i.e. $'one\ntwo\nthree\nfour'
while IFS= read -r id; do
((i++))
echo "restarting container $i: $id"
sudo docker start $id
# sudo docker exec -it $id python train.py --resume # single-GPU
sudo docker exec -d $id python utils/aws/resume.py # multi-scenario
done <<<"$list"
fi

161
app/yolov5/utils/benchmarks.py Normal file
View File

@ -0,0 +1,161 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Run YOLOv5 benchmarks on all supported export formats
Format | `export.py --include` | Model
--- | --- | ---
PyTorch | - | yolov5s.pt
TorchScript | `torchscript` | yolov5s.torchscript
ONNX | `onnx` | yolov5s.onnx
OpenVINO | `openvino` | yolov5s_openvino_model/
TensorRT | `engine` | yolov5s.engine
CoreML | `coreml` | yolov5s.mlmodel
TensorFlow SavedModel | `saved_model` | yolov5s_saved_model/
TensorFlow GraphDef | `pb` | yolov5s.pb
TensorFlow Lite | `tflite` | yolov5s.tflite
TensorFlow Edge TPU | `edgetpu` | yolov5s_edgetpu.tflite
TensorFlow.js | `tfjs` | yolov5s_web_model/
Requirements:
$ pip install -r requirements.txt coremltools onnx onnx-simplifier onnxruntime openvino-dev tensorflow-cpu # CPU
$ pip install -r requirements.txt coremltools onnx onnx-simplifier onnxruntime-gpu openvino-dev tensorflow # GPU
$ pip install -U nvidia-tensorrt --index-url https://pypi.ngc.nvidia.com # TensorRT
Usage:
$ python utils/benchmarks.py --weights yolov5s.pt --img 640
"""
import argparse
import platform
import sys
import time
from pathlib import Path
import pandas as pd
FILE = Path(__file__).resolve()
ROOT = FILE.parents[1] # YOLOv5 root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
# ROOT = ROOT.relative_to(Path.cwd()) # relative
import app.yolov5.export
import app.yolov5.val
from app.yolov5.utils import notebook_init
from app.yolov5.utils.general import LOGGER, check_yaml, file_size, print_args
from app.yolov5.utils.torch_utils import select_device
def run(
weights=ROOT / 'yolov5s.pt', # weights path
imgsz=640, # inference size (pixels)
batch_size=1, # batch size
data=ROOT / 'data/coco128.yaml', # dataset.yaml path
device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
half=False, # use FP16 half-precision inference
test=False, # test exports only
pt_only=False, # test PyTorch only
hard_fail=False, # throw error on benchmark failure
):
y, t = [], time.time()
device = select_device(device)
for i, (name, f, suffix, cpu, gpu) in export.export_formats().iterrows(): # index, (name, file, suffix, CPU, GPU)
try:
assert i not in (9, 10), 'inference not supported' # Edge TPU and TF.js are unsupported
assert i != 5 or platform.system() == 'Darwin', 'inference only supported on macOS>=10.13' # CoreML
if 'cpu' in device.type:
assert cpu, 'inference not supported on CPU'
if 'cuda' in device.type:
assert gpu, 'inference not supported on GPU'
# Export
if f == '-':
w = weights # PyTorch format
else:
w = export.run(weights=weights, imgsz=[imgsz], include=[f], device=device, half=half)[-1] # all others
assert suffix in str(w), 'export failed'
# Validate
result = val.run(data, w, batch_size, imgsz, plots=False, device=device, task='benchmark', half=half)
metrics = result[0] # metrics (mp, mr, map50, map, *losses(box, obj, cls))
speeds = result[2] # times (preprocess, inference, postprocess)
y.append([name, round(file_size(w), 1), round(metrics[3], 4), round(speeds[1], 2)]) # MB, mAP, t_inference
except Exception as e:
if hard_fail:
assert type(e) is AssertionError, f'Benchmark --hard-fail for {name}: {e}'
LOGGER.warning(f'WARNING: Benchmark failure for {name}: {e}')
y.append([name, None, None, None]) # mAP, t_inference
if pt_only and i == 0:
break # break after PyTorch
# Print results
LOGGER.info('\n')
parse_opt()
notebook_init() # print system info
c = ['Format', 'Size (MB)', 'mAP50-95', 'Inference time (ms)'] if map else ['Format', 'Export', '', '']
py = pd.DataFrame(y, columns=c)
LOGGER.info(f'\nBenchmarks complete ({time.time() - t:.2f}s)')
LOGGER.info(str(py if map else py.iloc[:, :2]))
if hard_fail and isinstance(hard_fail, str):
metrics = py['mAP50-95'].array # values to compare to floor
floor = eval(hard_fail) # minimum metric floor to pass, i.e. = 0.29 mAP for YOLOv5n
assert all(x > floor for x in metrics if pd.notna(x)), f'HARD FAIL: mAP50-95 < floor {floor}'
return py
def test(
weights=ROOT / 'yolov5s.pt', # weights path
imgsz=640, # inference size (pixels)
batch_size=1, # batch size
data=ROOT / 'data/coco128.yaml', # dataset.yaml path
device='', # cuda device, i.e. 0 or 0,1,2,3 or cpu
half=False, # use FP16 half-precision inference
test=False, # test exports only
pt_only=False, # test PyTorch only
hard_fail=False, # throw error on benchmark failure
):
y, t = [], time.time()
device = select_device(device)
for i, (name, f, suffix, gpu) in export.export_formats().iterrows(): # index, (name, file, suffix, gpu-capable)
try:
w = weights if f == '-' else \
export.run(weights=weights, imgsz=[imgsz], include=[f], device=device, half=half)[-1] # weights
assert suffix in str(w), 'export failed'
y.append([name, True])
except Exception:
y.append([name, False]) # mAP, t_inference
# Print results
LOGGER.info('\n')
parse_opt()
notebook_init() # print system info
py = pd.DataFrame(y, columns=['Format', 'Export'])
LOGGER.info(f'\nExports complete ({time.time() - t:.2f}s)')
LOGGER.info(str(py))
return py
def parse_opt():
parser = argparse.ArgumentParser()
parser.add_argument('--weights', type=str, default=ROOT / 'yolov5s.pt', help='weights path')
parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='inference size (pixels)')
parser.add_argument('--batch-size', type=int, default=1, help='batch size')
parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='dataset.yaml path')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference')
parser.add_argument('--test', action='store_true', help='test exports only')
parser.add_argument('--pt-only', action='store_true', help='test PyTorch only')
parser.add_argument('--hard-fail', nargs='?', const=True, default=False, help='Exception on error or < min metric')
opt = parser.parse_args()
opt.data = check_yaml(opt.data) # check YAML
print_args(vars(opt))
return opt
def main(opt):
test(**vars(opt)) if opt.test else run(**vars(opt))
if __name__ == "__main__":
opt = parse_opt()
main(opt)

76
app/yolov5/utils/callbacks.py Normal file
View File

@ -0,0 +1,76 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Callback utils
"""
import threading
class Callbacks:
""""
Handles all registered callbacks for YOLOv5 Hooks
"""
def __init__(self):
# Define the available callbacks
self._callbacks = {
'on_pretrain_routine_start': [],
'on_pretrain_routine_end': [],
'on_train_start': [],
'on_train_epoch_start': [],
'on_train_batch_start': [],
'optimizer_step': [],
'on_before_zero_grad': [],
'on_train_batch_end': [],
'on_train_epoch_end': [],
'on_val_start': [],
'on_val_batch_start': [],
'on_val_image_end': [],
'on_val_batch_end': [],
'on_val_end': [],
'on_fit_epoch_end': [], # fit = train + val
'on_model_save': [],
'on_train_end': [],
'on_params_update': [],
'teardown': [],}
self.stop_training = False # set True to interrupt training
def register_action(self, hook, name='', callback=None):
"""
Register a new action to a callback hook
Args:
hook: The callback hook name to register the action to
name: The name of the action for later reference
callback: The callback to fire
"""
assert hook in self._callbacks, f"hook '{hook}' not found in callbacks {self._callbacks}"
assert callable(callback), f"callback '{callback}' is not callable"
self._callbacks[hook].append({'name': name, 'callback': callback})
def get_registered_actions(self, hook=None):
""""
Returns all the registered actions by callback hook
Args:
hook: The name of the hook to check, defaults to all
"""
return self._callbacks[hook] if hook else self._callbacks
def run(self, hook, *args, thread=False, **kwargs):
"""
Loop through the registered actions and fire all callbacks on main thread
Args:
hook: The name of the hook to check, defaults to all
args: Arguments to receive from YOLOv5
thread: (boolean) Run callbacks in daemon thread
kwargs: Keyword Arguments to receive from YOLOv5
"""
assert hook in self._callbacks, f"hook '{hook}' not found in callbacks {self._callbacks}"
for logger in self._callbacks[hook]:
if thread:
threading.Thread(target=logger['callback'], args=args, kwargs=kwargs, daemon=True).start()
else:
logger['callback'](*args, **kwargs)

1163
app/yolov5/utils/dataloaders.py Normal file
View File

File diff suppressed because it is too large Load Diff

65
app/yolov5/utils/docker/Dockerfile Normal file
View File

@ -0,0 +1,65 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
# Builds ultralytics/yolov5:latest image on DockerHub https://hub.docker.com/r/ultralytics/yolov5
# Image is CUDA-optimized for YOLOv5 single/multi-GPU training and inference
# Start FROM NVIDIA PyTorch image https://ngc.nvidia.com/catalog/containers/nvidia:pytorch
FROM nvcr.io/nvidia/pytorch:22.07-py3
RUN rm -rf /opt/pytorch # remove 1.2GB dir
# Downloads to user config dir
ADD https://ultralytics.com/assets/Arial.ttf https://ultralytics.com/assets/Arial.Unicode.ttf /root/.config/Ultralytics/
# Install linux packages
RUN apt update && apt install --no-install-recommends -y zip htop screen libgl1-mesa-glx
# Install pip packages
COPY requirements.txt .
RUN python -m pip install --upgrade pip wheel
RUN pip uninstall -y Pillow torchtext torch torchvision
RUN pip install --no-cache -r requirements.txt albumentations wandb gsutil notebook Pillow>=9.1.0 \
'opencv-python<4.6.0.66' \
--extra-index-url https://download.pytorch.org/whl/cu113
# Create working directory
RUN mkdir -p /usr/src/app
WORKDIR /usr/src/app
# Copy contents
# COPY . /usr/src/app (issues as not a .git directory)
RUN git clone https://github.com/ultralytics/yolov5 /usr/src/app
# Set environment variables
ENV OMP_NUM_THREADS=8
# Usage Examples -------------------------------------------------------------------------------------------------------
# Build and Push
# t=ultralytics/yolov5:latest && sudo docker build -f utils/docker/Dockerfile -t $t . && sudo docker push $t
# Pull and Run
# t=ultralytics/yolov5:latest && sudo docker pull $t && sudo docker run -it --ipc=host --gpus all $t
# Pull and Run with local directory access
# t=ultralytics/yolov5:latest && sudo docker pull $t && sudo docker run -it --ipc=host --gpus all -v "$(pwd)"/datasets:/usr/src/datasets $t
# Kill all
# sudo docker kill $(sudo docker ps -q)
# Kill all image-based
# sudo docker kill $(sudo docker ps -qa --filter ancestor=ultralytics/yolov5:latest)
# DockerHub tag update
# t=ultralytics/yolov5:latest tnew=ultralytics/yolov5:v6.2 && sudo docker pull $t && sudo docker tag $t $tnew && sudo docker push $tnew
# Clean up
# docker system prune -a --volumes
# Update Ubuntu drivers
# https://www.maketecheasier.com/install-nvidia-drivers-ubuntu/
# DDP test
# python -m torch.distributed.run --nproc_per_node 2 --master_port 1 train.py --epochs 3
# GCP VM from Image
# docker.io/ultralytics/yolov5:latest

41
app/yolov5/utils/docker/Dockerfile-arm64 Normal file
View File

@ -0,0 +1,41 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
# Builds ultralytics/yolov5:latest-arm64 image on DockerHub https://hub.docker.com/r/ultralytics/yolov5
# Image is aarch64-compatible for Apple M1 and other ARM architectures i.e. Jetson Nano and Raspberry Pi
# Start FROM Ubuntu image https://hub.docker.com/_/ubuntu
FROM arm64v8/ubuntu:20.04
# Downloads to user config dir
ADD https://ultralytics.com/assets/Arial.ttf https://ultralytics.com/assets/Arial.Unicode.ttf /root/.config/Ultralytics/
# Install linux packages
RUN apt update
RUN DEBIAN_FRONTEND=noninteractive TZ=Etc/UTC apt install -y tzdata
RUN apt install --no-install-recommends -y python3-pip git zip curl htop gcc libgl1-mesa-glx libglib2.0-0 libpython3-dev
# RUN alias python=python3
# Install pip packages
COPY requirements.txt .
RUN python3 -m pip install --upgrade pip wheel
RUN pip install --no-cache -r requirements.txt gsutil notebook \
tensorflow-aarch64
# tensorflowjs \
# onnx onnx-simplifier onnxruntime \
# coremltools openvino-dev \
# Create working directory
RUN mkdir -p /usr/src/app
WORKDIR /usr/src/app
# Copy contents
# COPY . /usr/src/app (issues as not a .git directory)
RUN git clone https://github.com/ultralytics/yolov5 /usr/src/app
# Usage Examples -------------------------------------------------------------------------------------------------------
# Build and Push
# t=ultralytics/yolov5:latest-M1 && sudo docker build --platform linux/arm64 -f utils/docker/Dockerfile-arm64 -t $t . && sudo docker push $t
# Pull and Run
# t=ultralytics/yolov5:latest-M1 && sudo docker pull $t && sudo docker run -it --ipc=host -v "$(pwd)"/datasets:/usr/src/datasets $t

40
app/yolov5/utils/docker/Dockerfile-cpu Normal file
View File

@ -0,0 +1,40 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
# Builds ultralytics/yolov5:latest-cpu image on DockerHub https://hub.docker.com/r/ultralytics/yolov5
# Image is CPU-optimized for ONNX, OpenVINO and PyTorch YOLOv5 deployments
# Start FROM Ubuntu image https://hub.docker.com/_/ubuntu
FROM ubuntu:20.04
# Downloads to user config dir
ADD https://ultralytics.com/assets/Arial.ttf https://ultralytics.com/assets/Arial.Unicode.ttf /root/.config/Ultralytics/
# Install linux packages
RUN apt update
RUN DEBIAN_FRONTEND=noninteractive TZ=Etc/UTC apt install -y tzdata
RUN apt install --no-install-recommends -y python3-pip git zip curl htop libgl1-mesa-glx libglib2.0-0 libpython3-dev
# RUN alias python=python3
# Install pip packages
COPY requirements.txt .
RUN python3 -m pip install --upgrade pip wheel
RUN pip install --no-cache -r requirements.txt albumentations gsutil notebook \
coremltools onnx onnx-simplifier onnxruntime tensorflow-cpu tensorflowjs \
# openvino-dev \
--extra-index-url https://download.pytorch.org/whl/cpu
# Create working directory
RUN mkdir -p /usr/src/app
WORKDIR /usr/src/app
# Copy contents
# COPY . /usr/src/app (issues as not a .git directory)
RUN git clone https://github.com/ultralytics/yolov5 /usr/src/app
# Usage Examples -------------------------------------------------------------------------------------------------------
# Build and Push
# t=ultralytics/yolov5:latest-cpu && sudo docker build -f utils/docker/Dockerfile-cpu -t $t . && sudo docker push $t
# Pull and Run
# t=ultralytics/yolov5:latest-cpu && sudo docker pull $t && sudo docker run -it --ipc=host -v "$(pwd)"/datasets:/usr/src/datasets $t

192
app/yolov5/utils/downloads.py Normal file
View File

@ -0,0 +1,192 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Download utils
"""
import logging
import os
import platform
import subprocess
import time
import urllib
from pathlib import Path
from zipfile import ZipFile
import requests
import torch
def is_url(url, check_online=True):
# Check if online file exists
try:
url = str(url)
result = urllib.parse.urlparse(url)
assert all([result.scheme, result.netloc, result.path]) # check if is url
return (urllib.request.urlopen(url).getcode() == 200) if check_online else True # check if exists online
except (AssertionError, urllib.request.HTTPError):
return False
def gsutil_getsize(url=''):
# gs://bucket/file size https://cloud.google.com/storage/docs/gsutil/commands/du
s = subprocess.check_output(f'gsutil du {url}', shell=True).decode('utf-8')
return eval(s.split(' ')[0]) if len(s) else 0 # bytes
def url_getsize(url='https://ultralytics.com/images/bus.jpg'):
# Return downloadable file size in bytes
response = requests.head(url, allow_redirects=True)
return int(response.headers.get('content-length', -1))
def safe_download(file, url, url2=None, min_bytes=1E0, error_msg=''):
# Attempts to download file from url or url2, checks and removes incomplete downloads < min_bytes
from utils.general import LOGGER
file = Path(file)
assert_msg = f"Downloaded file '{file}' does not exist or size is < min_bytes={min_bytes}"
try: # url1
LOGGER.info(f'Downloading {url} to {file}...')
torch.hub.download_url_to_file(url, str(file), progress=LOGGER.level <= logging.INFO)
assert file.exists() and file.stat().st_size > min_bytes, assert_msg # check
except Exception as e: # url2
if file.exists():
file.unlink() # remove partial downloads
LOGGER.info(f'ERROR: {e}\nRe-attempting {url2 or url} to {file}...')
os.system(f"curl -# -L '{url2 or url}' -o '{file}' --retry 3 -C -") # curl download, retry and resume on fail
finally:
if not file.exists() or file.stat().st_size < min_bytes: # check
if file.exists():
file.unlink() # remove partial downloads
LOGGER.info(f"ERROR: {assert_msg}\n{error_msg}")
LOGGER.info('')
def attempt_download(file, repo='ultralytics/yolov5', release='v6.2'):
# Attempt file download from GitHub release assets if not found locally. release = 'latest', 'v6.2', etc.
from utils.general import LOGGER
def github_assets(repository, version='latest'):
# Return GitHub repo tag (i.e. 'v6.2') and assets (i.e. ['yolov5s.pt', 'yolov5m.pt', ...])
if version != 'latest':
version = f'tags/{version}' # i.e. tags/v6.2
response = requests.get(f'https://api.github.com/repos/{repository}/releases/{version}').json() # github api
return response['tag_name'], [x['name'] for x in response['assets']] # tag, assets
file = Path(str(file).strip().replace("'", ''))
if not file.exists():
# URL specified
name = Path(urllib.parse.unquote(str(file))).name # decode '%2F' to '/' etc.
if str(file).startswith(('http:/', 'https:/')): # download
url = str(file).replace(':/', '://') # Pathlib turns :// -> :/
file = name.split('?')[0] # parse authentication https://url.com/file.txt?auth...
if Path(file).is_file():
LOGGER.info(f'Found {url} locally at {file}') # file already exists
else:
safe_download(file=file, url=url, min_bytes=1E5)
return file
# GitHub assets
assets = [
'yolov5n.pt', 'yolov5s.pt', 'yolov5m.pt', 'yolov5l.pt', 'yolov5x.pt', 'yolov5n6.pt', 'yolov5s6.pt',
'yolov5m6.pt', 'yolov5l6.pt', 'yolov5x6.pt']
try:
tag, assets = github_assets(repo, release)
except Exception:
try:
tag, assets = github_assets(repo) # latest release
except Exception:
try:
tag = subprocess.check_output('git tag', shell=True, stderr=subprocess.STDOUT).decode().split()[-1]
except Exception:
tag = release
file.parent.mkdir(parents=True, exist_ok=True) # make parent dir (if required)
if name in assets:
url3 = 'https://drive.google.com/drive/folders/1EFQTEUeXWSFww0luse2jB9M1QNZQGwNl' # backup gdrive mirror
safe_download(
file,
url=f'https://github.com/{repo}/releases/download/{tag}/{name}',
url2=f'https://storage.googleapis.com/{repo}/{tag}/{name}', # backup url (optional)
min_bytes=1E5,
error_msg=f'{file} missing, try downloading from https://github.com/{repo}/releases/{tag} or {url3}')
return str(file)
def gdrive_download(id='16TiPfZj7htmTyhntwcZyEEAejOUxuT6m', file='tmp.zip'):
# Downloads a file from Google Drive. from yolov5.utils.downloads import *; gdrive_download()
t = time.time()
file = Path(file)
cookie = Path('cookie') # gdrive cookie
print(f'Downloading https://drive.google.com/uc?export=download&id={id} as {file}... ', end='')
if file.exists():
file.unlink() # remove existing file
if cookie.exists():
cookie.unlink() # remove existing cookie
# Attempt file download
out = "NUL" if platform.system() == "Windows" else "/dev/null"
os.system(f'curl -c ./cookie -s -L "drive.google.com/uc?export=download&id={id}" > {out}')
if os.path.exists('cookie'): # large file
s = f'curl -Lb ./cookie "drive.google.com/uc?export=download&confirm={get_token()}&id={id}" -o {file}'
else: # small file
s = f'curl -s -L -o {file} "drive.google.com/uc?export=download&id={id}"'
r = os.system(s) # execute, capture return
if cookie.exists():
cookie.unlink() # remove existing cookie
# Error check
if r != 0:
if file.exists():
file.unlink() # remove partial
print('Download error ') # raise Exception('Download error')
return r
# Unzip if archive
if file.suffix == '.zip':
print('unzipping... ', end='')
ZipFile(file).extractall(path=file.parent) # unzip
file.unlink() # remove zip
print(f'Done ({time.time() - t:.1f}s)')
return r
def get_token(cookie="./cookie"):
with open(cookie) as f:
for line in f:
if "download" in line:
return line.split()[-1]
return ""
# Google utils: https://cloud.google.com/storage/docs/reference/libraries ----------------------------------------------
#
#
# def upload_blob(bucket_name, source_file_name, destination_blob_name):
# # Uploads a file to a bucket
# # https://cloud.google.com/storage/docs/uploading-objects#storage-upload-object-python
#
# storage_client = storage.Client()
# bucket = storage_client.get_bucket(bucket_name)
# blob = bucket.blob(destination_blob_name)
#
# blob.upload_from_filename(source_file_name)
#
# print('File {} uploaded to {}.'.format(
# source_file_name,
# destination_blob_name))
#
#
# def download_blob(bucket_name, source_blob_name, destination_file_name):
# # Uploads a blob from a bucket
# storage_client = storage.Client()
# bucket = storage_client.get_bucket(bucket_name)
# blob = bucket.blob(source_blob_name)
#
# blob.download_to_filename(destination_file_name)
#
# print('Blob {} downloaded to {}.'.format(
# source_blob_name,
# destination_file_name))

73
app/yolov5/utils/flask_rest_api/README.md Normal file
View File

@ -0,0 +1,73 @@
# Flask REST API
[REST](https://en.wikipedia.org/wiki/Representational_state_transfer) [API](https://en.wikipedia.org/wiki/API)s are
commonly used to expose Machine Learning (ML) models to other services. This folder contains an example REST API
created using Flask to expose the YOLOv5s model from [PyTorch Hub](https://pytorch.org/hub/ultralytics_yolov5/).
## Requirements
[Flask](https://palletsprojects.com/p/flask/) is required. Install with:
```shell
$ pip install Flask
```
## Run
After Flask installation run:
```shell
$ python3 restapi.py --port 5000
```
Then use [curl](https://curl.se/) to perform a request:
```shell
$ curl -X POST -F image=@zidane.jpg 'http://localhost:5000/v1/object-detection/yolov5s'
```
The model inference results are returned as a JSON response:
```json
[
{
"class": 0,
"confidence": 0.8900438547,
"height": 0.9318675399,
"name": "person",
"width": 0.3264600933,
"xcenter": 0.7438579798,
"ycenter": 0.5207948685
},
{
"class": 0,
"confidence": 0.8440024257,
"height": 0.7155083418,
"name": "person",
"width": 0.6546785235,
"xcenter": 0.427829951,
"ycenter": 0.6334488392
},
{
"class": 27,
"confidence": 0.3771208823,
"height": 0.3902671337,
"name": "tie",
"width": 0.0696444362,
"xcenter": 0.3675483763,
"ycenter": 0.7991207838
},
{
"class": 27,
"confidence": 0.3527112305,
"height": 0.1540903747,
"name": "tie",
"width": 0.0336618312,
"xcenter": 0.7814827561,
"ycenter": 0.5065554976
}
]
```
An example python script to perform inference using [requests](https://docs.python-requests.org/en/master/) is given
in `example_request.py`

19
app/yolov5/utils/flask_rest_api/example_request.py Normal file
View File

@ -0,0 +1,19 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Perform test request
"""
import pprint
import requests
DETECTION_URL = "http://localhost:5000/v1/object-detection/yolov5s"
IMAGE = "zidane.jpg"
# Read image
with open(IMAGE, "rb") as f:
image_data = f.read()
response = requests.post(DETECTION_URL, files={"image": image_data}).json()
pprint.pprint(response)

48
app/yolov5/utils/flask_rest_api/restapi.py Normal file
View File

@ -0,0 +1,48 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Run a Flask REST API exposing one or more YOLOv5s models
"""
import argparse
import io
import torch
from flask import Flask, request
from PIL import Image
app = Flask(__name__)
models = {}
DETECTION_URL = "/v1/object-detection/<model>"
@app.route(DETECTION_URL, methods=["POST"])
def predict(model):
if request.method != "POST":
return
if request.files.get("image"):
# Method 1
# with request.files["image"] as f:
# im = Image.open(io.BytesIO(f.read()))
# Method 2
im_file = request.files["image"]
im_bytes = im_file.read()
im = Image.open(io.BytesIO(im_bytes))
if model in models:
results = models[model](im, size=640) # reduce size=320 for faster inference
return results.pandas().xyxy[0].to_json(orient="records")
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Flask API exposing YOLOv5 model")
parser.add_argument("--port", default=5000, type=int, help="port number")
parser.add_argument('--model', nargs='+', default=['yolov5s'], help='model(s) to run, i.e. --model yolov5n yolov5s')
opt = parser.parse_args()
for m in opt.model:
models[m] = torch.hub.load("ultralytics/yolov5", m, force_reload=True, skip_validation=True)
app.run(host="0.0.0.0", port=opt.port) # debug=True causes Restarting with stat

1046
app/yolov5/utils/general.py Normal file
View File

File diff suppressed because it is too large Load Diff

25
app/yolov5/utils/google_app_engine/Dockerfile Normal file
View File

@ -0,0 +1,25 @@
FROM gcr.io/google-appengine/python
# Create a virtualenv for dependencies. This isolates these packages from
# system-level packages.
# Use -p python3 or -p python3.7 to select python version. Default is version 2.
RUN virtualenv /env -p python3
# Setting these environment variables are the same as running
# source /env/bin/activate.
ENV VIRTUAL_ENV /env
ENV PATH /env/bin:$PATH
RUN apt-get update && apt-get install -y python-opencv
# Copy the application's requirements.txt and run pip to install all
# dependencies into the virtualenv.
ADD requirements.txt /app/requirements.txt
RUN pip install -r /app/requirements.txt
# Add the application source code.
ADD . /app
# Run a WSGI server to serve the application. gunicorn must be declared as
# a dependency in requirements.txt.
CMD gunicorn -b :$PORT main:app

4
app/yolov5/utils/google_app_engine/additional_requirements.txt Normal file
View File

@ -0,0 +1,4 @@
# add these requirements in your app on top of the existing ones
pip==21.1
Flask==1.0.2
gunicorn==19.9.0

14
app/yolov5/utils/google_app_engine/app.yaml Normal file
View File

@ -0,0 +1,14 @@
runtime: custom
env: flex
service: yolov5app
liveness_check:
initial_delay_sec: 600
manual_scaling:
instances: 1
resources:
cpu: 1
memory_gb: 4
disk_size_gb: 20

339
app/yolov5/utils/loggers/__init__.py Normal file
View File

@ -0,0 +1,339 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Logging utils
"""
import os
import warnings
from pathlib import Path
import pkg_resources as pkg
import torch
from torch.utils.tensorboard import SummaryWriter
from app.yolov5.utils.general import colorstr, cv2
from app.yolov5.utils.loggers.clearml.clearml_utils import ClearmlLogger
from app.yolov5.utils.loggers.wandb.wandb_utils import WandbLogger
from app.yolov5.utils.plots import plot_images, plot_labels, plot_results
from app.yolov5.utils.torch_utils import de_parallel
LOGGERS = ('csv', 'tb', 'wandb', 'clearml') # *.csv, TensorBoard, Weights & Biases, ClearML
RANK = int(os.getenv('RANK', -1))
try:
import wandb
assert hasattr(wandb, '__version__') # verify package import not local dir
if pkg.parse_version(wandb.__version__) >= pkg.parse_version('0.12.2') and RANK in {0, -1}:
try:
wandb_login_success = wandb.login(timeout=30)
except wandb.errors.UsageError: # known non-TTY terminal issue
wandb_login_success = False
if not wandb_login_success:
wandb = None
except (ImportError, AssertionError):
wandb = None
try:
import clearml
assert hasattr(clearml, '__version__') # verify package import not local dir
except (ImportError, AssertionError):
clearml = None
class Loggers():
# YOLOv5 Loggers class
def __init__(self, save_dir=None, weights=None, opt=None, hyp=None, logger=None, include=LOGGERS):
self.save_dir = save_dir
self.weights = weights
self.opt = opt
self.hyp = hyp
self.plots = not opt.noplots # plot results
self.logger = logger # for printing results to console
self.include = include
self.keys = [
'train/box_loss',
'train/obj_loss',
'train/cls_loss', # train loss
'metrics/precision',
'metrics/recall',
'metrics/mAP_0.5',
'metrics/mAP_0.5:0.95', # metrics
'val/box_loss',
'val/obj_loss',
'val/cls_loss', # val loss
'x/lr0',
'x/lr1',
'x/lr2'] # params
self.best_keys = ['best/epoch', 'best/precision', 'best/recall', 'best/mAP_0.5', 'best/mAP_0.5:0.95']
for k in LOGGERS:
setattr(self, k, None) # init empty logger dictionary
self.csv = True # always log to csv
# Messages
if not wandb:
prefix = colorstr('Weights & Biases: ')
s = f"{prefix}run 'pip install wandb' to automatically track and visualize YOLOv5 🚀 runs in Weights & Biases"
self.logger.info(s)
if not clearml:
prefix = colorstr('ClearML: ')
s = f"{prefix}run 'pip install clearml' to automatically track, visualize and remotely train YOLOv5 🚀 in ClearML"
self.logger.info(s)
# TensorBoard
s = self.save_dir
if 'tb' in self.include and not self.opt.evolve:
prefix = colorstr('TensorBoard: ')
self.logger.info(f"{prefix}Start with 'tensorboard --logdir {s.parent}', view at http://localhost:6006/")
self.tb = SummaryWriter(str(s))
# W&B
if wandb and 'wandb' in self.include:
wandb_artifact_resume = isinstance(self.opt.resume, str) and self.opt.resume.startswith('wandb-artifact://')
run_id = torch.load(self.weights).get('wandb_id') if self.opt.resume and not wandb_artifact_resume else None
self.opt.hyp = self.hyp # add hyperparameters
self.wandb = WandbLogger(self.opt, run_id)
# temp warn. because nested artifacts not supported after 0.12.10
if pkg.parse_version(wandb.__version__) >= pkg.parse_version('0.12.11'):
s = "YOLOv5 temporarily requires wandb version 0.12.10 or below. Some features may not work as expected."
self.logger.warning(s)
else:
self.wandb = None
# ClearML
if clearml and 'clearml' in self.include:
self.clearml = ClearmlLogger(self.opt, self.hyp)
else:
self.clearml = None
@property
def remote_dataset(self):
# Get data_dict if custom dataset artifact link is provided
data_dict = None
if self.clearml:
data_dict = self.clearml.data_dict
if self.wandb:
data_dict = self.wandb.data_dict
return data_dict
def on_train_start(self):
# Callback runs on train start
pass
def on_pretrain_routine_end(self, labels, names):
# Callback runs on pre-train routine end
if self.plots:
plot_labels(labels, names, self.save_dir)
paths = self.save_dir.glob('*labels*.jpg') # training labels
if self.wandb:
self.wandb.log({"Labels": [wandb.Image(str(x), caption=x.name) for x in paths]})
# if self.clearml:
# pass # ClearML saves these images automatically using hooks
def on_train_batch_end(self, model, ni, imgs, targets, paths):
# Callback runs on train batch end
# ni: number integrated batches (since train start)
if self.plots:
if ni < 3:
f = self.save_dir / f'train_batch{ni}.jpg' # filename
plot_images(imgs, targets, paths, f)
if ni == 0 and self.tb and not self.opt.sync_bn:
log_tensorboard_graph(self.tb, model, imgsz=(self.opt.imgsz, self.opt.imgsz))
if ni == 10 and (self.wandb or self.clearml):
files = sorted(self.save_dir.glob('train*.jpg'))
if self.wandb:
self.wandb.log({'Mosaics': [wandb.Image(str(f), caption=f.name) for f in files if f.exists()]})
if self.clearml:
self.clearml.log_debug_samples(files, title='Mosaics')
def on_train_epoch_end(self, epoch):
# Callback runs on train epoch end
if self.wandb:
self.wandb.current_epoch = epoch + 1
def on_val_image_end(self, pred, predn, path, names, im):
# Callback runs on val image end
if self.wandb:
self.wandb.val_one_image(pred, predn, path, names, im)
if self.clearml:
self.clearml.log_image_with_boxes(path, pred, names, im)
def on_val_end(self):
# Callback runs on val end
if self.wandb or self.clearml:
files = sorted(self.save_dir.glob('val*.jpg'))
if self.wandb:
self.wandb.log({"Validation": [wandb.Image(str(f), caption=f.name) for f in files]})
if self.clearml:
self.clearml.log_debug_samples(files, title='Validation')
def on_fit_epoch_end(self, vals, epoch, best_fitness, fi):
# Callback runs at the end of each fit (train+val) epoch
x = dict(zip(self.keys, vals))
if self.csv:
file = self.save_dir / 'results.csv'
n = len(x) + 1 # number of cols
s = '' if file.exists() else (('%20s,' * n % tuple(['epoch'] + self.keys)).rstrip(',') + '\n') # add header
with open(file, 'a') as f:
f.write(s + ('%20.5g,' * n % tuple([epoch] + vals)).rstrip(',') + '\n')
if self.tb:
for k, v in x.items():
self.tb.add_scalar(k, v, epoch)
elif self.clearml: # log to ClearML if TensorBoard not used
for k, v in x.items():
title, series = k.split('/')
self.clearml.task.get_logger().report_scalar(title, series, v, epoch)
if self.wandb:
if best_fitness == fi:
best_results = [epoch] + vals[3:7]
for i, name in enumerate(self.best_keys):
self.wandb.wandb_run.summary[name] = best_results[i] # log best results in the summary
self.wandb.log(x)
self.wandb.end_epoch(best_result=best_fitness == fi)
if self.clearml:
self.clearml.current_epoch_logged_images = set() # reset epoch image limit
self.clearml.current_epoch += 1
def on_model_save(self, last, epoch, final_epoch, best_fitness, fi):
# Callback runs on model save event
if (epoch + 1) % self.opt.save_period == 0 and not final_epoch and self.opt.save_period != -1:
if self.wandb:
self.wandb.log_model(last.parent, self.opt, epoch, fi, best_model=best_fitness == fi)
if self.clearml:
self.clearml.task.update_output_model(model_path=str(last),
model_name='Latest Model',
auto_delete_file=False)
def on_train_end(self, last, best, epoch, results):
# Callback runs on training end, i.e. saving best model
if self.plots:
plot_results(file=self.save_dir / 'results.csv') # save results.png
files = ['results.png', 'confusion_matrix.png', *(f'{x}_curve.png' for x in ('F1', 'PR', 'P', 'R'))]
files = [(self.save_dir / f) for f in files if (self.save_dir / f).exists()] # filter
self.logger.info(f"Results saved to {colorstr('bold', self.save_dir)}")
if self.tb and not self.clearml: # These images are already captured by ClearML by now, we don't want doubles
for f in files:
self.tb.add_image(f.stem, cv2.imread(str(f))[..., ::-1], epoch, dataformats='HWC')
if self.wandb:
self.wandb.log(dict(zip(self.keys[3:10], results)))
self.wandb.log({"Results": [wandb.Image(str(f), caption=f.name) for f in files]})
# Calling wandb.log. TODO: Refactor this into WandbLogger.log_model
if not self.opt.evolve:
wandb.log_artifact(str(best if best.exists() else last),
type='model',
name=f'run_{self.wandb.wandb_run.id}_model',
aliases=['latest', 'best', 'stripped'])
self.wandb.finish_run()
if self.clearml and not self.opt.evolve:
self.clearml.task.update_output_model(model_path=str(best if best.exists() else last),
name='Best Model',
auto_delete_file=False)
def on_params_update(self, params: dict):
# Update hyperparams or configs of the experiment
if self.wandb:
self.wandb.wandb_run.config.update(params, allow_val_change=True)
class GenericLogger:
"""
YOLOv5 General purpose logger for non-task specific logging
Usage: from utils.loggers import GenericLogger; logger = GenericLogger(...)
Arguments
opt: Run arguments
console_logger: Console logger
include: loggers to include
"""
def __init__(self, opt, console_logger, include=('tb', 'wandb')):
# init default loggers
self.save_dir = Path(opt.save_dir)
self.include = include
self.console_logger = console_logger
self.csv = self.save_dir / 'results.csv' # CSV logger
if 'tb' in self.include:
prefix = colorstr('TensorBoard: ')
self.console_logger.info(
f"{prefix}Start with 'tensorboard --logdir {self.save_dir.parent}', view at http://localhost:6006/")
self.tb = SummaryWriter(str(self.save_dir))
if wandb and 'wandb' in self.include:
self.wandb = wandb.init(project=web_project_name(str(opt.project)),
name=None if opt.name == "exp" else opt.name,
config=opt)
else:
self.wandb = None
def log_metrics(self, metrics, epoch):
# Log metrics dictionary to all loggers
if self.csv:
keys, vals = list(metrics.keys()), list(metrics.values())
n = len(metrics) + 1 # number of cols
s = '' if self.csv.exists() else (('%23s,' * n % tuple(['epoch'] + keys)).rstrip(',') + '\n') # header
with open(self.csv, 'a') as f:
f.write(s + ('%23.5g,' * n % tuple([epoch] + vals)).rstrip(',') + '\n')
if self.tb:
for k, v in metrics.items():
self.tb.add_scalar(k, v, epoch)
if self.wandb:
self.wandb.log(metrics, step=epoch)
def log_images(self, files, name='Images', epoch=0):
# Log images to all loggers
files = [Path(f) for f in (files if isinstance(files, (tuple, list)) else [files])] # to Path
files = [f for f in files if f.exists()] # filter by exists
if self.tb:
for f in files:
self.tb.add_image(f.stem, cv2.imread(str(f))[..., ::-1], epoch, dataformats='HWC')
if self.wandb:
self.wandb.log({name: [wandb.Image(str(f), caption=f.name) for f in files]}, step=epoch)
def log_graph(self, model, imgsz=(640, 640)):
# Log model graph to all loggers
if self.tb:
log_tensorboard_graph(self.tb, model, imgsz)
def log_model(self, model_path, epoch=0, metadata={}):
# Log model to all loggers
if self.wandb:
art = wandb.Artifact(name=f"run_{wandb.run.id}_model", type="model", metadata=metadata)
art.add_file(str(model_path))
wandb.log_artifact(art)
def update_params(self, params):
# Update the paramters logged
if self.wandb:
wandb.run.config.update(params, allow_val_change=True)
def log_tensorboard_graph(tb, model, imgsz=(640, 640)):
# Log model graph to TensorBoard
try:
p = next(model.parameters()) # for device, type
imgsz = (imgsz, imgsz) if isinstance(imgsz, int) else imgsz # expand
im = torch.zeros((1, 3, *imgsz)).to(p.device).type_as(p) # input image (WARNING: must be zeros, not empty)
with warnings.catch_warnings():
warnings.simplefilter('ignore') # suppress jit trace warning
tb.add_graph(torch.jit.trace(de_parallel(model), im, strict=False), [])
except Exception as e:
print(f'WARNING: TensorBoard graph visualization failure {e}')
def web_project_name(project):
# Convert local project name to web project name
if not project.startswith('runs/train'):
return project
suffix = '-Classify' if project.endswith('-cls') else '-Segment' if project.endswith('-seg') else ''
return f'YOLOv5{suffix}'

222
app/yolov5/utils/loggers/clearml/README.md Normal file
View File

@ -0,0 +1,222 @@
# ClearML Integration
<img align="center" src="https://github.com/thepycoder/clearml_screenshots/raw/main/logos_dark.png#gh-light-mode-only" alt="Clear|ML"><img align="center" src="https://github.com/thepycoder/clearml_screenshots/raw/main/logos_light.png#gh-dark-mode-only" alt="Clear|ML">
## About ClearML
[ClearML](https://cutt.ly/yolov5-tutorial-clearml) is an [open-source](https://github.com/allegroai/clearml) toolbox designed to save you time ⏱️.
🔨 Track every YOLOv5 training run in the <b>experiment manager</b>
🔧 Version and easily access your custom training data with the integrated ClearML <b>Data Versioning Tool</b>
🔦 <b>Remotely train and monitor</b> your YOLOv5 training runs using ClearML Agent
🔬 Get the very best mAP using ClearML <b>Hyperparameter Optimization</b>
🔭 Turn your newly trained <b>YOLOv5 model into an API</b> with just a few commands using ClearML Serving
<br />
And so much more. It's up to you how many of these tools you want to use, you can stick to the experiment manager, or chain them all together into an impressive pipeline!
<br />
<br />
![ClearML scalars dashboard](https://github.com/thepycoder/clearml_screenshots/raw/main/experiment_manager_with_compare.gif)
<br />
<br />
## 🦾 Setting Things Up
To keep track of your experiments and/or data, ClearML needs to communicate to a server. You have 2 options to get one:
Either sign up for free to the [ClearML Hosted Service](https://cutt.ly/yolov5-tutorial-clearml) or you can set up your own server, see [here](https://clear.ml/docs/latest/docs/deploying_clearml/clearml_server). Even the server is open-source, so even if you're dealing with sensitive data, you should be good to go!
1. Install the `clearml` python package:
```bash
pip install clearml
```
1. Connect the ClearML SDK to the server by [creating credentials](https://app.clear.ml/settings/workspace-configuration) (go right top to Settings -> Workspace -> Create new credentials), then execute the command below and follow the instructions:
```bash
clearml-init
```
That's it! You're done 😎
<br />
## 🚀 Training YOLOv5 With ClearML
To enable ClearML experiment tracking, simply install the ClearML pip package.
```bash
pip install clearml
```
This will enable integration with the YOLOv5 training script. Every training run from now on, will be captured and stored by the ClearML experiment manager. If you want to change the `project_name` or `task_name`, head over to our custom logger, where you can change it: `utils/loggers/clearml/clearml_utils.py`
```bash
python train.py --img 640 --batch 16 --epochs 3 --data coco128.yaml --weights yolov5s.pt --cache
```
This will capture:
- Source code + uncommitted changes
- Installed packages
- (Hyper)parameters
- Model files (use `--save-period n` to save a checkpoint every n epochs)
- Console output
- Scalars (mAP_0.5, mAP_0.5:0.95, precision, recall, losses, learning rates, ...)
- General info such as machine details, runtime, creation date etc.
- All produced plots such as label correlogram and confusion matrix
- Images with bounding boxes per epoch
- Mosaic per epoch
- Validation images per epoch
- ...
That's a lot right? 🤯
Now, we can visualize all of this information in the ClearML UI to get an overview of our training progress. Add custom columns to the table view (such as e.g. mAP_0.5) so you can easily sort on the best performing model. Or select multiple experiments and directly compare them!
There even more we can do with all of this information, like hyperparameter optimization and remote execution, so keep reading if you want to see how that works!
<br />
## 🔗 Dataset Version Management
Versioning your data separately from your code is generally a good idea and makes it easy to aqcuire the latest version too. This repository supports supplying a dataset version ID and it will make sure to get the data if it's not there yet. Next to that, this workflow also saves the used dataset ID as part of the task parameters, so you will always know for sure which data was used in which experiment!
![ClearML Dataset Interface](https://github.com/thepycoder/clearml_screenshots/raw/main/clearml_data.gif)
### Prepare Your Dataset
The YOLOv5 repository supports a number of different datasets by using yaml files containing their information. By default datasets are downloaded to the `../datasets` folder in relation to the repository root folder. So if you downloaded the `coco128` dataset using the link in the yaml or with the scripts provided by yolov5, you get this folder structure:
```
..
|_ yolov5
|_ datasets
|_ coco128
|_ images
|_ labels
|_ LICENSE
|_ README.txt
```
But this can be any dataset you wish. Feel free to use your own, as long as you keep to this folder structure.
Next, ⚠️**copy the corresponding yaml file to the root of the dataset folder**⚠️. This yaml files contains the information ClearML will need to properly use the dataset. You can make this yourself too, of course, just follow the structure of the example yamls.
Basically we need the following keys: `path`, `train`, `test`, `val`, `nc`, `names`.
```
..
|_ yolov5
|_ datasets
|_ coco128
|_ images
|_ labels
|_ coco128.yaml # <---- HERE!
|_ LICENSE
|_ README.txt
```
### Upload Your Dataset
To get this dataset into ClearML as a versionned dataset, go to the dataset root folder and run the following command:
```bash
cd coco128
clearml-data sync --project YOLOv5 --name coco128 --folder .
```
The command `clearml-data sync` is actually a shorthand command. You could also run these commands one after the other:
```bash
# Optionally add --parent <parent_dataset_id> if you want to base
# this version on another dataset version, so no duplicate files are uploaded!
clearml-data create --name coco128 --project YOLOv5
clearml-data add --files .
clearml-data close
```
### Run Training Using A ClearML Dataset
Now that you have a ClearML dataset, you can very simply use it to train custom YOLOv5 🚀 models!
```bash
python train.py --img 640 --batch 16 --epochs 3 --data clearml://<your_dataset_id> --weights yolov5s.pt --cache
```
<br />
## 👀 Hyperparameter Optimization
Now that we have our experiments and data versioned, it's time to take a look at what we can build on top!
Using the code information, installed packages and environment details, the experiment itself is now **completely reproducible**. In fact, ClearML allows you to clone an experiment and even change its parameters. We can then just rerun it with these new parameters automatically, this is basically what HPO does!
To **run hyperparameter optimization locally**, we've included a pre-made script for you. Just make sure a training task has been run at least once, so it is in the ClearML experiment manager, we will essentially clone it and change its hyperparameters.
You'll need to fill in the ID of this `template task` in the script found at `utils/loggers/clearml/hpo.py` and then just run it :) You can change `task.execute_locally()` to `task.execute()` to put it in a ClearML queue and have a remote agent work on it instead.
```bash
# To use optuna, install it first, otherwise you can change the optimizer to just be RandomSearch
pip install optuna
python utils/loggers/clearml/hpo.py
```
![HPO](https://github.com/thepycoder/clearml_screenshots/raw/main/hpo.png)
## 🤯 Remote Execution (advanced)
Running HPO locally is really handy, but what if we want to run our experiments on a remote machine instead? Maybe you have access to a very powerful GPU machine on-site or you have some budget to use cloud GPUs.
This is where the ClearML Agent comes into play. Check out what the agent can do here:
- [YouTube video](https://youtu.be/MX3BrXnaULs)
- [Documentation](https://clear.ml/docs/latest/docs/clearml_agent)
In short: every experiment tracked by the experiment manager contains enough information to reproduce it on a different machine (installed packages, uncommitted changes etc.). So a ClearML agent does just that: it listens to a queue for incoming tasks and when it finds one, it recreates the environment and runs it while still reporting scalars, plots etc. to the experiment manager.
You can turn any machine (a cloud VM, a local GPU machine, your own laptop ... ) into a ClearML agent by simply running:
```bash
clearml-agent daemon --queue <queues_to_listen_to> [--docker]
```
### Cloning, Editing And Enqueuing
With our agent running, we can give it some work. Remember from the HPO section that we can clone a task and edit the hyperparameters? We can do that from the interface too!
🪄 Clone the experiment by right clicking it
🎯 Edit the hyperparameters to what you wish them to be
⏳ Enqueue the task to any of the queues by right clicking it
![Enqueue a task from the UI](https://github.com/thepycoder/clearml_screenshots/raw/main/enqueue.gif)
### Executing A Task Remotely
Now you can clone a task like we explained above, or simply mark your current script by adding `task.execute_remotely()` and on execution it will be put into a queue, for the agent to start working on!
To run the YOLOv5 training script remotely, all you have to do is add this line to the training.py script after the clearml logger has been instatiated:
```python
# ...
# Loggers
data_dict = None
if RANK in {-1, 0}:
loggers = Loggers(save_dir, weights, opt, hyp, LOGGER) # loggers instance
if loggers.clearml:
loggers.clearml.task.execute_remotely(queue='my_queue') # <------ ADD THIS LINE
# Data_dict is either None is user did not choose for ClearML dataset or is filled in by ClearML
data_dict = loggers.clearml.data_dict
# ...
```
When running the training script after this change, python will run the script up until that line, after which it will package the code and send it to the queue instead!
### Autoscaling workers
ClearML comes with autoscalers too! This tool will automatically spin up new remote machines in the cloud of your choice (AWS, GCP, Azure) and turn them into ClearML agents for you whenever there are experiments detected in the queue. Once the tasks are processed, the autoscaler will automatically shut down the remote machines and you stop paying!
Check out the autoscalers getting started video below.
[![Watch the video](https://img.youtube.com/vi/j4XVMAaUt3E/0.jpg)](https://youtu.be/j4XVMAaUt3E)

0
app/yolov5/utils/loggers/clearml/__init__.py Normal file
View File

156
app/yolov5/utils/loggers/clearml/clearml_utils.py Normal file
View File

@ -0,0 +1,156 @@
"""Main Logger class for ClearML experiment tracking."""
import glob
import re
from pathlib import Path
import numpy as np
import yaml
from app.yolov5.utils.plots import Annotator, colors
try:
import clearml
from clearml import Dataset, Task
assert hasattr(clearml, '__version__') # verify package import not local dir
except (ImportError, AssertionError):
clearml = None
def construct_dataset(clearml_info_string):
"""Load in a clearml dataset and fill the internal data_dict with its contents.
"""
dataset_id = clearml_info_string.replace('clearml://', '')
dataset = Dataset.get(dataset_id=dataset_id)
dataset_root_path = Path(dataset.get_local_copy())
# We'll search for the yaml file definition in the dataset
yaml_filenames = list(glob.glob(str(dataset_root_path / "*.yaml")) + glob.glob(str(dataset_root_path / "*.yml")))
if len(yaml_filenames) > 1:
raise ValueError('More than one yaml file was found in the dataset root, cannot determine which one contains '
'the dataset definition this way.')
elif len(yaml_filenames) == 0:
raise ValueError('No yaml definition found in dataset root path, check that there is a correct yaml file '
'inside the dataset root path.')
with open(yaml_filenames[0]) as f:
dataset_definition = yaml.safe_load(f)
assert set(dataset_definition.keys()).issuperset(
{'train', 'test', 'val', 'nc', 'names'}
), "The right keys were not found in the yaml file, make sure it at least has the following keys: ('train', 'test', 'val', 'nc', 'names')"
data_dict = dict()
data_dict['train'] = str(
(dataset_root_path / dataset_definition['train']).resolve()) if dataset_definition['train'] else None
data_dict['test'] = str(
(dataset_root_path / dataset_definition['test']).resolve()) if dataset_definition['test'] else None
data_dict['val'] = str(
(dataset_root_path / dataset_definition['val']).resolve()) if dataset_definition['val'] else None
data_dict['nc'] = dataset_definition['nc']
data_dict['names'] = dataset_definition['names']
return data_dict
class ClearmlLogger:
"""Log training runs, datasets, models, and predictions to ClearML.
This logger sends information to ClearML at app.clear.ml or to your own hosted server. By default,
this information includes hyperparameters, system configuration and metrics, model metrics, code information and
basic data metrics and analyses.
By providing additional command line arguments to train.py, datasets,
models and predictions can also be logged.
"""
def __init__(self, opt, hyp):
"""
- Initialize ClearML Task, this object will capture the experiment
- Upload dataset version to ClearML Data if opt.upload_dataset is True
arguments:
opt (namespace) -- Commandline arguments for this run
hyp (dict) -- Hyperparameters for this run
"""
self.current_epoch = 0
# Keep tracked of amount of logged images to enforce a limit
self.current_epoch_logged_images = set()
# Maximum number of images to log to clearML per epoch
self.max_imgs_to_log_per_epoch = 16
# Get the interval of epochs when bounding box images should be logged
self.bbox_interval = opt.bbox_interval
self.clearml = clearml
self.task = None
self.data_dict = None
if self.clearml:
self.task = Task.init(
project_name='YOLOv5',
task_name='training',
tags=['YOLOv5'],
output_uri=True,
auto_connect_frameworks={'pytorch': False}
# We disconnect pytorch auto-detection, because we added manual model save points in the code
)
# ClearML's hooks will already grab all general parameters
# Only the hyperparameters coming from the yaml config file
# will have to be added manually!
self.task.connect(hyp, name='Hyperparameters')
# Get ClearML Dataset Version if requested
if opt.data.startswith('clearml://'):
# data_dict should have the following keys:
# names, nc (number of classes), test, train, val (all three relative paths to ../datasets)
self.data_dict = construct_dataset(opt.data)
# Set data to data_dict because wandb will crash without this information and opt is the best way
# to give it to them
opt.data = self.data_dict
def log_debug_samples(self, files, title='Debug Samples'):
"""
Log files (images) as debug samples in the ClearML task.
arguments:
files (List(PosixPath)) a list of file paths in PosixPath format
title (str) A title that groups together images with the same values
"""
for f in files:
if f.exists():
it = re.search(r'_batch(\d+)', f.name)
iteration = int(it.groups()[0]) if it else 0
self.task.get_logger().report_image(title=title,
series=f.name.replace(it.group(), ''),
local_path=str(f),
iteration=iteration)
def log_image_with_boxes(self, image_path, boxes, class_names, image, conf_threshold=0.25):
"""
Draw the bounding boxes on a single image and report the result as a ClearML debug sample.
arguments:
image_path (PosixPath) the path the original image file
boxes (list): list of scaled predictions in the format - [xmin, ymin, xmax, ymax, confidence, class]
class_names (dict): dict containing mapping of class int to class name
image (Tensor): A torch tensor containing the actual image data
"""
if len(self.current_epoch_logged_images) < self.max_imgs_to_log_per_epoch and self.current_epoch >= 0:
# Log every bbox_interval times and deduplicate for any intermittend extra eval runs
if self.current_epoch % self.bbox_interval == 0 and image_path not in self.current_epoch_logged_images:
im = np.ascontiguousarray(np.moveaxis(image.mul(255).clamp(0, 255).byte().cpu().numpy(), 0, 2))
annotator = Annotator(im=im, pil=True)
for i, (conf, class_nr, box) in enumerate(zip(boxes[:, 4], boxes[:, 5], boxes[:, :4])):
color = colors(i)
class_name = class_names[int(class_nr)]
confidence_percentage = round(float(conf) * 100, 2)
label = f"{class_name}: {confidence_percentage}%"
if conf > conf_threshold:
annotator.rectangle(box.cpu().numpy(), outline=color)
annotator.box_label(box.cpu().numpy(), label=label, color=color)
annotated_image = annotator.result()
self.task.get_logger().report_image(title='Bounding Boxes',
series=image_path.name,
iteration=self.current_epoch,
image=annotated_image)
self.current_epoch_logged_images.add(image_path)

84
app/yolov5/utils/loggers/clearml/hpo.py Normal file
View File

@ -0,0 +1,84 @@
from clearml import Task
# Connecting ClearML with the current process,
# from here on everything is logged automatically
from clearml.automation import HyperParameterOptimizer, UniformParameterRange
from clearml.automation.optuna import OptimizerOptuna
task = Task.init(project_name='Hyper-Parameter Optimization',
task_name='YOLOv5',
task_type=Task.TaskTypes.optimizer,
reuse_last_task_id=False)
# Example use case:
optimizer = HyperParameterOptimizer(
# This is the experiment we want to optimize
base_task_id='<your_template_task_id>',
# here we define the hyper-parameters to optimize
# Notice: The parameter name should exactly match what you see in the UI: <section_name>/<parameter>
# For Example, here we see in the base experiment a section Named: "General"
# under it a parameter named "batch_size", this becomes "General/batch_size"
# If you have `argparse` for example, then arguments will appear under the "Args" section,
# and you should instead pass "Args/batch_size"
hyper_parameters=[
UniformParameterRange('Hyperparameters/lr0', min_value=1e-5, max_value=1e-1),
UniformParameterRange('Hyperparameters/lrf', min_value=0.01, max_value=1.0),
UniformParameterRange('Hyperparameters/momentum', min_value=0.6, max_value=0.98),
UniformParameterRange('Hyperparameters/weight_decay', min_value=0.0, max_value=0.001),
UniformParameterRange('Hyperparameters/warmup_epochs', min_value=0.0, max_value=5.0),
UniformParameterRange('Hyperparameters/warmup_momentum', min_value=0.0, max_value=0.95),
UniformParameterRange('Hyperparameters/warmup_bias_lr', min_value=0.0, max_value=0.2),
UniformParameterRange('Hyperparameters/box', min_value=0.02, max_value=0.2),
UniformParameterRange('Hyperparameters/cls', min_value=0.2, max_value=4.0),
UniformParameterRange('Hyperparameters/cls_pw', min_value=0.5, max_value=2.0),
UniformParameterRange('Hyperparameters/obj', min_value=0.2, max_value=4.0),
UniformParameterRange('Hyperparameters/obj_pw', min_value=0.5, max_value=2.0),
UniformParameterRange('Hyperparameters/iou_t', min_value=0.1, max_value=0.7),
UniformParameterRange('Hyperparameters/anchor_t', min_value=2.0, max_value=8.0),
UniformParameterRange('Hyperparameters/fl_gamma', min_value=0.0, max_value=4.0),
UniformParameterRange('Hyperparameters/hsv_h', min_value=0.0, max_value=0.1),
UniformParameterRange('Hyperparameters/hsv_s', min_value=0.0, max_value=0.9),
UniformParameterRange('Hyperparameters/hsv_v', min_value=0.0, max_value=0.9),
UniformParameterRange('Hyperparameters/degrees', min_value=0.0, max_value=45.0),
UniformParameterRange('Hyperparameters/translate', min_value=0.0, max_value=0.9),
UniformParameterRange('Hyperparameters/scale', min_value=0.0, max_value=0.9),
UniformParameterRange('Hyperparameters/shear', min_value=0.0, max_value=10.0),
UniformParameterRange('Hyperparameters/perspective', min_value=0.0, max_value=0.001),
UniformParameterRange('Hyperparameters/flipud', min_value=0.0, max_value=1.0),
UniformParameterRange('Hyperparameters/fliplr', min_value=0.0, max_value=1.0),
UniformParameterRange('Hyperparameters/mosaic', min_value=0.0, max_value=1.0),
UniformParameterRange('Hyperparameters/mixup', min_value=0.0, max_value=1.0),
UniformParameterRange('Hyperparameters/copy_paste', min_value=0.0, max_value=1.0)],
# this is the objective metric we want to maximize/minimize
objective_metric_title='metrics',
objective_metric_series='mAP_0.5',
# now we decide if we want to maximize it or minimize it (accuracy we maximize)
objective_metric_sign='max',
# let us limit the number of concurrent experiments,
# this in turn will make sure we do dont bombard the scheduler with experiments.
# if we have an auto-scaler connected, this, by proxy, will limit the number of machine
max_number_of_concurrent_tasks=1,
# this is the optimizer class (actually doing the optimization)
# Currently, we can choose from GridSearch, RandomSearch or OptimizerBOHB (Bayesian optimization Hyper-Band)
optimizer_class=OptimizerOptuna,
# If specified only the top K performing Tasks will be kept, the others will be automatically archived
save_top_k_tasks_only=5, # 5,
compute_time_limit=None,
total_max_jobs=20,
min_iteration_per_job=None,
max_iteration_per_job=None,
)
# report every 10 seconds, this is way too often, but we are testing here
optimizer.set_report_period(10 / 60)
# You can also use the line below instead to run all the optimizer tasks locally, without using queues or agent
# an_optimizer.start_locally(job_complete_callback=job_complete_callback)
# set the time limit for the optimization process (2 hours)
optimizer.set_time_limit(in_minutes=120.0)
# Start the optimization process in the local environment
optimizer.start_locally()
# wait until process is done (notice we are controlling the optimization process in the background)
optimizer.wait()
# make sure background optimization stopped
optimizer.stop()
print('We are done, good bye')

162
app/yolov5/utils/loggers/wandb/README.md Normal file
View File

@ -0,0 +1,162 @@
📚 This guide explains how to use **Weights & Biases** (W&B) with YOLOv5 🚀. UPDATED 29 September 2021.
- [About Weights & Biases](#about-weights-&-biases)
- [First-Time Setup](#first-time-setup)
- [Viewing runs](#viewing-runs)
- [Disabling wandb](#disabling-wandb)
- [Advanced Usage: Dataset Versioning and Evaluation](#advanced-usage)
- [Reports: Share your work with the world!](#reports)
## About Weights & Biases
Think of [W&B](https://wandb.ai/site?utm_campaign=repo_yolo_wandbtutorial) like GitHub for machine learning models. With a few lines of code, save everything you need to debug, compare and reproduce your models — architecture, hyperparameters, git commits, model weights, GPU usage, and even datasets and predictions.
Used by top researchers including teams at OpenAI, Lyft, Github, and MILA, W&B is part of the new standard of best practices for machine learning. How W&B can help you optimize your machine learning workflows:
- [Debug](https://wandb.ai/wandb/getting-started/reports/Visualize-Debug-Machine-Learning-Models--VmlldzoyNzY5MDk#Free-2) model performance in real time
- [GPU usage](https://wandb.ai/wandb/getting-started/reports/Visualize-Debug-Machine-Learning-Models--VmlldzoyNzY5MDk#System-4) visualized automatically
- [Custom charts](https://wandb.ai/wandb/customizable-charts/reports/Powerful-Custom-Charts-To-Debug-Model-Peformance--VmlldzoyNzY4ODI) for powerful, extensible visualization
- [Share insights](https://wandb.ai/wandb/getting-started/reports/Visualize-Debug-Machine-Learning-Models--VmlldzoyNzY5MDk#Share-8) interactively with collaborators
- [Optimize hyperparameters](https://docs.wandb.com/sweeps) efficiently
- [Track](https://docs.wandb.com/artifacts) datasets, pipelines, and production models
## First-Time Setup
<details open>
<summary> Toggle Details </summary>
When you first train, W&B will prompt you to create a new account and will generate an **API key** for you. If you are an existing user you can retrieve your key from https://wandb.ai/authorize. This key is used to tell W&B where to log your data. You only need to supply your key once, and then it is remembered on the same device.
W&B will create a cloud **project** (default is 'YOLOv5') for your training runs, and each new training run will be provided a unique run **name** within that project as project/name. You can also manually set your project and run name as:
```shell
$ python train.py --project ... --name ...
```
YOLOv5 notebook example: <a href="https://colab.research.google.com/github/ultralytics/yolov5/blob/master/tutorial.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"></a> <a href="https://www.kaggle.com/ultralytics/yolov5"><img src="https://kaggle.com/static/images/open-in-kaggle.svg" alt="Open In Kaggle"></a>
<img width="960" alt="Screen Shot 2021-09-29 at 10 23 13 PM" src="https://user-images.githubusercontent.com/26833433/135392431-1ab7920a-c49d-450a-b0b0-0c86ec86100e.png">
</details>
## Viewing Runs
<details open>
<summary> Toggle Details </summary>
Run information streams from your environment to the W&B cloud console as you train. This allows you to monitor and even cancel runs in <b>realtime</b> . All important information is logged:
- Training & Validation losses
- Metrics: Precision, Recall, mAP@0.5, mAP@0.5:0.95
- Learning Rate over time
- A bounding box debugging panel, showing the training progress over time
- GPU: Type, **GPU Utilization**, power, temperature, **CUDA memory usage**
- System: Disk I/0, CPU utilization, RAM memory usage
- Your trained model as W&B Artifact
- Environment: OS and Python types, Git repository and state, **training command**
<p align="center"><img width="900" alt="Weights & Biases dashboard" src="https://user-images.githubusercontent.com/26833433/135390767-c28b050f-8455-4004-adb0-3b730386e2b2.png"></p>
</details>
## Disabling wandb
- training after running `wandb disabled` inside that directory creates no wandb run
![Screenshot (84)](https://user-images.githubusercontent.com/15766192/143441777-c780bdd7-7cb4-4404-9559-b4316030a985.png)
- To enable wandb again, run `wandb online`
![Screenshot (85)](https://user-images.githubusercontent.com/15766192/143441866-7191b2cb-22f0-4e0f-ae64-2dc47dc13078.png)
## Advanced Usage
You can leverage W&B artifacts and Tables integration to easily visualize and manage your datasets, models and training evaluations. Here are some quick examples to get you started.
<details open>
<h3> 1: Train and Log Evaluation simultaneousy </h3>
This is an extension of the previous section, but it'll also training after uploading the dataset. <b> This also evaluation Table</b>
Evaluation table compares your predictions and ground truths across the validation set for each epoch. It uses the references to the already uploaded datasets,
so no images will be uploaded from your system more than once.
<details open>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python train.py --upload_data val</code>
![Screenshot from 2021-11-21 17-40-06](https://user-images.githubusercontent.com/15766192/142761183-c1696d8c-3f38-45ab-991a-bb0dfd98ae7d.png)
</details>
<h3>2. Visualize and Version Datasets</h3>
Log, visualize, dynamically query, and understand your data with <a href='https://docs.wandb.ai/guides/data-vis/tables'>W&B Tables</a>. You can use the following command to log your dataset as a W&B Table. This will generate a <code>{dataset}_wandb.yaml</code> file which can be used to train from dataset artifact.
<details>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python utils/logger/wandb/log_dataset.py --project ... --name ... --data .. </code>
![Screenshot (64)](https://user-images.githubusercontent.com/15766192/128486078-d8433890-98a3-4d12-8986-b6c0e3fc64b9.png)
</details>
<h3> 3: Train using dataset artifact </h3>
When you upload a dataset as described in the first section, you get a new config file with an added `_wandb` to its name. This file contains the information that
can be used to train a model directly from the dataset artifact. <b> This also logs evaluation </b>
<details>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python train.py --data {data}_wandb.yaml </code>
![Screenshot (72)](https://user-images.githubusercontent.com/15766192/128979739-4cf63aeb-a76f-483f-8861-1c0100b938a5.png)
</details>
<h3> 4: Save model checkpoints as artifacts </h3>
To enable saving and versioning checkpoints of your experiment, pass `--save_period n` with the base cammand, where `n` represents checkpoint interval.
You can also log both the dataset and model checkpoints simultaneously. If not passed, only the final model will be logged
<details>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python train.py --save_period 1 </code>
![Screenshot (68)](https://user-images.githubusercontent.com/15766192/128726138-ec6c1f60-639d-437d-b4ee-3acd9de47ef3.png)
</details>
</details>
<h3> 5: Resume runs from checkpoint artifacts. </h3>
Any run can be resumed using artifacts if the <code>--resume</code> argument starts with <code>wandb-artifact://</code> prefix followed by the run path, i.e, <code>wandb-artifact://username/project/runid </code>. This doesn't require the model checkpoint to be present on the local system.
<details>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python train.py --resume wandb-artifact://{run_path} </code>
![Screenshot (70)](https://user-images.githubusercontent.com/15766192/128728988-4e84b355-6c87-41ae-a591-14aecf45343e.png)
</details>
<h3> 6: Resume runs from dataset artifact & checkpoint artifacts. </h3>
<b> Local dataset or model checkpoints are not required. This can be used to resume runs directly on a different device </b>
The syntax is same as the previous section, but you'll need to lof both the dataset and model checkpoints as artifacts, i.e, set bot <code>--upload_dataset</code> or
train from <code>_wandb.yaml</code> file and set <code>--save_period</code>
<details>
<summary> <b>Usage</b> </summary>
<b>Code</b> <code> $ python train.py --resume wandb-artifact://{run_path} </code>
![Screenshot (70)](https://user-images.githubusercontent.com/15766192/128728988-4e84b355-6c87-41ae-a591-14aecf45343e.png)
</details>
</details>
<h3> Reports </h3>
W&B Reports can be created from your saved runs for sharing online. Once a report is created you will receive a link you can use to publically share your results. Here is an example report created from the COCO128 tutorial trainings of all four YOLOv5 models ([link](https://wandb.ai/glenn-jocher/yolov5_tutorial/reports/YOLOv5-COCO128-Tutorial-Results--VmlldzozMDI5OTY)).
<img width="900" alt="Weights & Biases Reports" src="https://user-images.githubusercontent.com/26833433/135394029-a17eaf86-c6c1-4b1d-bb80-b90e83aaffa7.png">
## Environments
YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including [CUDA](https://developer.nvidia.com/cuda)/[CUDNN](https://developer.nvidia.com/cudnn), [Python](https://www.python.org/) and [PyTorch](https://pytorch.org/) preinstalled):
- **Google Colab and Kaggle** notebooks with free GPU: <a href="https://colab.research.google.com/github/ultralytics/yolov5/blob/master/tutorial.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"></a> <a href="https://www.kaggle.com/ultralytics/yolov5"><img src="https://kaggle.com/static/images/open-in-kaggle.svg" alt="Open In Kaggle"></a>
- **Google Cloud** Deep Learning VM. See [GCP Quickstart Guide](https://github.com/ultralytics/yolov5/wiki/GCP-Quickstart)
- **Amazon** Deep Learning AMI. See [AWS Quickstart Guide](https://github.com/ultralytics/yolov5/wiki/AWS-Quickstart)
- **Docker Image**. See [Docker Quickstart Guide](https://github.com/ultralytics/yolov5/wiki/Docker-Quickstart) <a href="https://hub.docker.com/r/ultralytics/yolov5"><img src="https://img.shields.io/docker/pulls/ultralytics/yolov5?logo=docker" alt="Docker Pulls"></a>
## Status
![CI CPU testing](https://github.com/ultralytics/yolov5/workflows/CI%20CPU%20testing/badge.svg)
If this badge is green, all [YOLOv5 GitHub Actions](https://github.com/ultralytics/yolov5/actions) Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training ([train.py](https://github.com/ultralytics/yolov5/blob/master/train.py)), validation ([val.py](https://github.com/ultralytics/yolov5/blob/master/val.py)), inference ([detect.py](https://github.com/ultralytics/yolov5/blob/master/detect.py)) and export ([export.py](https://github.com/ultralytics/yolov5/blob/master/export.py)) on macOS, Windows, and Ubuntu every 24 hours and on every commit.

0
app/yolov5/utils/loggers/wandb/__init__.py Normal file
View File

27
app/yolov5/utils/loggers/wandb/log_dataset.py Normal file
View File

@ -0,0 +1,27 @@
import argparse
from wandb_utils import WandbLogger
from utils.general import LOGGER
WANDB_ARTIFACT_PREFIX = 'wandb-artifact://'
def create_dataset_artifact(opt):
logger = WandbLogger(opt, None, job_type='Dataset Creation') # TODO: return value unused
if not logger.wandb:
LOGGER.info("install wandb using `pip install wandb` to log the dataset")
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--data', type=str, default='data/coco128.yaml', help='data.yaml path')
parser.add_argument('--single-cls', action='store_true', help='train as single-class dataset')
parser.add_argument('--project', type=str, default='YOLOv5', help='name of W&B Project')
parser.add_argument('--entity', default=None, help='W&B entity')
parser.add_argument('--name', type=str, default='log dataset', help='name of W&B run')
opt = parser.parse_args()
opt.resume = False # Explicitly disallow resume check for dataset upload job
create_dataset_artifact(opt)

41
app/yolov5/utils/loggers/wandb/sweep.py Normal file
View File

@ -0,0 +1,41 @@
import sys
from pathlib import Path
import wandb
FILE = Path(__file__).resolve()
ROOT = FILE.parents[3] # YOLOv5 root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
from train import parse_opt, train
from utils.callbacks import Callbacks
from utils.general import increment_path
from utils.torch_utils import select_device
def sweep():
wandb.init()
# Get hyp dict from sweep agent. Copy because train() modifies parameters which confused wandb.
hyp_dict = vars(wandb.config).get("_items").copy()
# Workaround: get necessary opt args
opt = parse_opt(known=True)
opt.batch_size = hyp_dict.get("batch_size")
opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok or opt.evolve))
opt.epochs = hyp_dict.get("epochs")
opt.nosave = True
opt.data = hyp_dict.get("data")
opt.weights = str(opt.weights)
opt.cfg = str(opt.cfg)
opt.data = str(opt.data)
opt.hyp = str(opt.hyp)
opt.project = str(opt.project)
device = select_device(opt.device, batch_size=opt.batch_size)
# train
train(hyp_dict, opt, device, callbacks=Callbacks())
if __name__ == "__main__":
sweep()

143
app/yolov5/utils/loggers/wandb/sweep.yaml Normal file
View File

@ -0,0 +1,143 @@
# Hyperparameters for training
# To set range-
# Provide min and max values as:
# parameter:
#
# min: scalar
# max: scalar
# OR
#
# Set a specific list of search space-
# parameter:
# values: [scalar1, scalar2, scalar3...]
#
# You can use grid, bayesian and hyperopt search strategy
# For more info on configuring sweeps visit - https://docs.wandb.ai/guides/sweeps/configuration
program: utils/loggers/wandb/sweep.py
method: random
metric:
name: metrics/mAP_0.5
goal: maximize
parameters:
# hyperparameters: set either min, max range or values list
data:
value: "data/coco128.yaml"
batch_size:
values: [64]
epochs:
values: [10]
lr0:
distribution: uniform
min: 1e-5
max: 1e-1
lrf:
distribution: uniform
min: 0.01
max: 1.0
momentum:
distribution: uniform
min: 0.6
max: 0.98
weight_decay:
distribution: uniform
min: 0.0
max: 0.001
warmup_epochs:
distribution: uniform
min: 0.0
max: 5.0
warmup_momentum:
distribution: uniform
min: 0.0
max: 0.95
warmup_bias_lr:
distribution: uniform
min: 0.0
max: 0.2
box:
distribution: uniform
min: 0.02
max: 0.2
cls:
distribution: uniform
min: 0.2
max: 4.0
cls_pw:
distribution: uniform
min: 0.5
max: 2.0
obj:
distribution: uniform
min: 0.2
max: 4.0
obj_pw:
distribution: uniform
min: 0.5
max: 2.0
iou_t:
distribution: uniform
min: 0.1
max: 0.7
anchor_t:
distribution: uniform
min: 2.0
max: 8.0
fl_gamma:
distribution: uniform
min: 0.0
max: 4.0
hsv_h:
distribution: uniform
min: 0.0
max: 0.1
hsv_s:
distribution: uniform
min: 0.0
max: 0.9
hsv_v:
distribution: uniform
min: 0.0
max: 0.9
degrees:
distribution: uniform
min: 0.0
max: 45.0
translate:
distribution: uniform
min: 0.0
max: 0.9
scale:
distribution: uniform
min: 0.0
max: 0.9
shear:
distribution: uniform
min: 0.0
max: 10.0
perspective:
distribution: uniform
min: 0.0
max: 0.001
flipud:
distribution: uniform
min: 0.0
max: 1.0
fliplr:
distribution: uniform
min: 0.0
max: 1.0
mosaic:
distribution: uniform
min: 0.0
max: 1.0
mixup:
distribution: uniform
min: 0.0
max: 1.0
copy_paste:
distribution: uniform
min: 0.0
max: 1.0

584
app/yolov5/utils/loggers/wandb/wandb_utils.py Normal file
View File

@ -0,0 +1,584 @@
"""Utilities and tools for tracking runs with Weights & Biases."""
import logging
import os
import sys
from contextlib import contextmanager
from pathlib import Path
from typing import Dict
import yaml
from tqdm import tqdm
FILE = Path(__file__).resolve()
ROOT = FILE.parents[3] # YOLOv5 root directory
if str(ROOT) not in sys.path:
sys.path.append(str(ROOT)) # add ROOT to PATH
from app.yolov5.utils.dataloaders import LoadImagesAndLabels, img2label_paths
from app.yolov5.utils.general import LOGGER, check_dataset, check_file
try:
import wandb
assert hasattr(wandb, '__version__') # verify package import not local dir
except (ImportError, AssertionError):
wandb = None
RANK = int(os.getenv('RANK', -1))
WANDB_ARTIFACT_PREFIX = 'wandb-artifact://'
def remove_prefix(from_string, prefix=WANDB_ARTIFACT_PREFIX):
return from_string[len(prefix):]
def check_wandb_config_file(data_config_file):
wandb_config = '_wandb.'.join(data_config_file.rsplit('.', 1)) # updated data.yaml path
if Path(wandb_config).is_file():
return wandb_config
return data_config_file
def check_wandb_dataset(data_file):
is_trainset_wandb_artifact = False
is_valset_wandb_artifact = False
if isinstance(data_file, dict):
# In that case another dataset manager has already processed it and we don't have to
return data_file
if check_file(data_file) and data_file.endswith('.yaml'):
with open(data_file, errors='ignore') as f:
data_dict = yaml.safe_load(f)
is_trainset_wandb_artifact = isinstance(data_dict['train'],
str) and data_dict['train'].startswith(WANDB_ARTIFACT_PREFIX)
is_valset_wandb_artifact = isinstance(data_dict['val'],
str) and data_dict['val'].startswith(WANDB_ARTIFACT_PREFIX)
if is_trainset_wandb_artifact or is_valset_wandb_artifact:
return data_dict
else:
return check_dataset(data_file)
def get_run_info(run_path):
run_path = Path(remove_prefix(run_path, WANDB_ARTIFACT_PREFIX))
run_id = run_path.stem
project = run_path.parent.stem
entity = run_path.parent.parent.stem
model_artifact_name = 'run_' + run_id + '_model'
return entity, project, run_id, model_artifact_name
def check_wandb_resume(opt):
process_wandb_config_ddp_mode(opt) if RANK not in [-1, 0] else None
if isinstance(opt.resume, str):
if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
if RANK not in [-1, 0]: # For resuming DDP runs
entity, project, run_id, model_artifact_name = get_run_info(opt.resume)
api = wandb.Api()
artifact = api.artifact(entity + '/' + project + '/' + model_artifact_name + ':latest')
modeldir = artifact.download()
opt.weights = str(Path(modeldir) / "last.pt")
return True
return None
def process_wandb_config_ddp_mode(opt):
with open(check_file(opt.data), errors='ignore') as f:
data_dict = yaml.safe_load(f) # data dict
train_dir, val_dir = None, None
if isinstance(data_dict['train'], str) and data_dict['train'].startswith(WANDB_ARTIFACT_PREFIX):
api = wandb.Api()
train_artifact = api.artifact(remove_prefix(data_dict['train']) + ':' + opt.artifact_alias)
train_dir = train_artifact.download()
train_path = Path(train_dir) / 'data/images/'
data_dict['train'] = str(train_path)
if isinstance(data_dict['val'], str) and data_dict['val'].startswith(WANDB_ARTIFACT_PREFIX):
api = wandb.Api()
val_artifact = api.artifact(remove_prefix(data_dict['val']) + ':' + opt.artifact_alias)
val_dir = val_artifact.download()
val_path = Path(val_dir) / 'data/images/'
data_dict['val'] = str(val_path)
if train_dir or val_dir:
ddp_data_path = str(Path(val_dir) / 'wandb_local_data.yaml')
with open(ddp_data_path, 'w') as f:
yaml.safe_dump(data_dict, f)
opt.data = ddp_data_path
class WandbLogger():
"""Log training runs, datasets, models, and predictions to Weights & Biases.
This logger sends information to W&B at wandb.ai. By default, this information
includes hyperparameters, system configuration and metrics, model metrics,
and basic data metrics and analyses.
By providing additional command line arguments to train.py, datasets,
models and predictions can also be logged.
For more on how this logger is used, see the Weights & Biases documentation:
https://docs.wandb.com/guides/integrations/yolov5
"""
def __init__(self, opt, run_id=None, job_type='Training'):
"""
- Initialize WandbLogger instance
- Upload dataset if opt.upload_dataset is True
- Setup training processes if job_type is 'Training'
arguments:
opt (namespace) -- Commandline arguments for this run
run_id (str) -- Run ID of W&B run to be resumed
job_type (str) -- To set the job_type for this run
"""
# Pre-training routine --
self.job_type = job_type
self.wandb, self.wandb_run = wandb, None if not wandb else wandb.run
self.val_artifact, self.train_artifact = None, None
self.train_artifact_path, self.val_artifact_path = None, None
self.result_artifact = None
self.val_table, self.result_table = None, None
self.bbox_media_panel_images = []
self.val_table_path_map = None
self.max_imgs_to_log = 16
self.wandb_artifact_data_dict = None
self.data_dict = None
# It's more elegant to stick to 1 wandb.init call,
# but useful config data is overwritten in the WandbLogger's wandb.init call
if isinstance(opt.resume, str): # checks resume from artifact
if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
entity, project, run_id, model_artifact_name = get_run_info(opt.resume)
model_artifact_name = WANDB_ARTIFACT_PREFIX + model_artifact_name
assert wandb, 'install wandb to resume wandb runs'
# Resume wandb-artifact:// runs here| workaround for not overwriting wandb.config
self.wandb_run = wandb.init(id=run_id,
project=project,
entity=entity,
resume='allow',
allow_val_change=True)
opt.resume = model_artifact_name
elif self.wandb:
self.wandb_run = wandb.init(config=opt,
resume="allow",
project='YOLOv5' if opt.project == 'runs/train' else Path(opt.project).stem,
entity=opt.entity,
name=opt.name if opt.name != 'exp' else None,
job_type=job_type,
id=run_id,
allow_val_change=True) if not wandb.run else wandb.run
if self.wandb_run:
if self.job_type == 'Training':
if opt.upload_dataset:
if not opt.resume:
self.wandb_artifact_data_dict = self.check_and_upload_dataset(opt)
if isinstance(opt.data, dict):
# This means another dataset manager has already processed the dataset info (e.g. ClearML)
# and they will have stored the already processed dict in opt.data
self.data_dict = opt.data
elif opt.resume:
# resume from artifact
if isinstance(opt.resume, str) and opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
self.data_dict = dict(self.wandb_run.config.data_dict)
else: # local resume
self.data_dict = check_wandb_dataset(opt.data)
else:
self.data_dict = check_wandb_dataset(opt.data)
self.wandb_artifact_data_dict = self.wandb_artifact_data_dict or self.data_dict
# write data_dict to config. useful for resuming from artifacts. Do this only when not resuming.
self.wandb_run.config.update({'data_dict': self.wandb_artifact_data_dict}, allow_val_change=True)
self.setup_training(opt)
if self.job_type == 'Dataset Creation':
self.wandb_run.config.update({"upload_dataset": True})
self.data_dict = self.check_and_upload_dataset(opt)
def check_and_upload_dataset(self, opt):
"""
Check if the dataset format is compatible and upload it as W&B artifact
arguments:
opt (namespace)-- Commandline arguments for current run
returns:
Updated dataset info dictionary where local dataset paths are replaced by WAND_ARFACT_PREFIX links.
"""
assert wandb, 'Install wandb to upload dataset'
config_path = self.log_dataset_artifact(opt.data, opt.single_cls,
'YOLOv5' if opt.project == 'runs/train' else Path(opt.project).stem)
with open(config_path, errors='ignore') as f:
wandb_data_dict = yaml.safe_load(f)
return wandb_data_dict
def setup_training(self, opt):
"""
Setup the necessary processes for training YOLO models:
- Attempt to download model checkpoint and dataset artifacts if opt.resume stats with WANDB_ARTIFACT_PREFIX
- Update data_dict, to contain info of previous run if resumed and the paths of dataset artifact if downloaded
- Setup log_dict, initialize bbox_interval
arguments:
opt (namespace) -- commandline arguments for this run
"""
self.log_dict, self.current_epoch = {}, 0
self.bbox_interval = opt.bbox_interval
if isinstance(opt.resume, str):
modeldir, _ = self.download_model_artifact(opt)
if modeldir:
self.weights = Path(modeldir) / "last.pt"
config = self.wandb_run.config
opt.weights, opt.save_period, opt.batch_size, opt.bbox_interval, opt.epochs, opt.hyp, opt.imgsz = str(
self.weights), config.save_period, config.batch_size, config.bbox_interval, config.epochs,\
config.hyp, config.imgsz
data_dict = self.data_dict
if self.val_artifact is None: # If --upload_dataset is set, use the existing artifact, don't download
self.train_artifact_path, self.train_artifact = self.download_dataset_artifact(
data_dict.get('train'), opt.artifact_alias)
self.val_artifact_path, self.val_artifact = self.download_dataset_artifact(
data_dict.get('val'), opt.artifact_alias)
if self.train_artifact_path is not None:
train_path = Path(self.train_artifact_path) / 'data/images/'
data_dict['train'] = str(train_path)
if self.val_artifact_path is not None:
val_path = Path(self.val_artifact_path) / 'data/images/'
data_dict['val'] = str(val_path)
if self.val_artifact is not None:
self.result_artifact = wandb.Artifact("run_" + wandb.run.id + "_progress", "evaluation")
columns = ["epoch", "id", "ground truth", "prediction"]
columns.extend(self.data_dict['names'])
self.result_table = wandb.Table(columns)
self.val_table = self.val_artifact.get("val")
if self.val_table_path_map is None:
self.map_val_table_path()
if opt.bbox_interval == -1:
self.bbox_interval = opt.bbox_interval = (opt.epochs // 10) if opt.epochs > 10 else 1
if opt.evolve or opt.noplots:
self.bbox_interval = opt.bbox_interval = opt.epochs + 1 # disable bbox_interval
train_from_artifact = self.train_artifact_path is not None and self.val_artifact_path is not None
# Update the the data_dict to point to local artifacts dir
if train_from_artifact:
self.data_dict = data_dict
def download_dataset_artifact(self, path, alias):
"""
download the model checkpoint artifact if the path starts with WANDB_ARTIFACT_PREFIX
arguments:
path -- path of the dataset to be used for training
alias (str)-- alias of the artifact to be download/used for training
returns:
(str, wandb.Artifact) -- path of the downladed dataset and it's corresponding artifact object if dataset
is found otherwise returns (None, None)
"""
if isinstance(path, str) and path.startswith(WANDB_ARTIFACT_PREFIX):
artifact_path = Path(remove_prefix(path, WANDB_ARTIFACT_PREFIX) + ":" + alias)
dataset_artifact = wandb.use_artifact(artifact_path.as_posix().replace("\\", "/"))
assert dataset_artifact is not None, "'Error: W&B dataset artifact doesn\'t exist'"
datadir = dataset_artifact.download()
return datadir, dataset_artifact
return None, None
def download_model_artifact(self, opt):
"""
download the model checkpoint artifact if the resume path starts with WANDB_ARTIFACT_PREFIX
arguments:
opt (namespace) -- Commandline arguments for this run
"""
if opt.resume.startswith(WANDB_ARTIFACT_PREFIX):
model_artifact = wandb.use_artifact(remove_prefix(opt.resume, WANDB_ARTIFACT_PREFIX) + ":latest")
assert model_artifact is not None, 'Error: W&B model artifact doesn\'t exist'
modeldir = model_artifact.download()
# epochs_trained = model_artifact.metadata.get('epochs_trained')
total_epochs = model_artifact.metadata.get('total_epochs')
is_finished = total_epochs is None
assert not is_finished, 'training is finished, can only resume incomplete runs.'
return modeldir, model_artifact
return None, None
def log_model(self, path, opt, epoch, fitness_score, best_model=False):
"""
Log the model checkpoint as W&B artifact
arguments:
path (Path) -- Path of directory containing the checkpoints
opt (namespace) -- Command line arguments for this run
epoch (int) -- Current epoch number
fitness_score (float) -- fitness score for current epoch
best_model (boolean) -- Boolean representing if the current checkpoint is the best yet.
"""
model_artifact = wandb.Artifact('run_' + wandb.run.id + '_model',
type='model',
metadata={
'original_url': str(path),
'epochs_trained': epoch + 1,
'save period': opt.save_period,
'project': opt.project,
'total_epochs': opt.epochs,
'fitness_score': fitness_score})
model_artifact.add_file(str(path / 'last.pt'), name='last.pt')
wandb.log_artifact(model_artifact,
aliases=['latest', 'last', 'epoch ' + str(self.current_epoch), 'best' if best_model else ''])
LOGGER.info(f"Saving model artifact on epoch {epoch + 1}")
def log_dataset_artifact(self, data_file, single_cls, project, overwrite_config=False):
"""
Log the dataset as W&B artifact and return the new data file with W&B links
arguments:
data_file (str) -- the .yaml file with information about the dataset like - path, classes etc.
single_class (boolean) -- train multi-class data as single-class
project (str) -- project name. Used to construct the artifact path
overwrite_config (boolean) -- overwrites the data.yaml file if set to true otherwise creates a new
file with _wandb postfix. Eg -> data_wandb.yaml
returns:
the new .yaml file with artifact links. it can be used to start training directly from artifacts
"""
upload_dataset = self.wandb_run.config.upload_dataset
log_val_only = isinstance(upload_dataset, str) and upload_dataset == 'val'
self.data_dict = check_dataset(data_file) # parse and check
data = dict(self.data_dict)
nc, names = (1, ['item']) if single_cls else (int(data['nc']), data['names'])
names = {k: v for k, v in enumerate(names)} # to index dictionary
# log train set
if not log_val_only:
self.train_artifact = self.create_dataset_table(LoadImagesAndLabels(data['train'], rect=True, batch_size=1),
names,
name='train') if data.get('train') else None
if data.get('train'):
data['train'] = WANDB_ARTIFACT_PREFIX + str(Path(project) / 'train')
self.val_artifact = self.create_dataset_table(
LoadImagesAndLabels(data['val'], rect=True, batch_size=1), names, name='val') if data.get('val') else None
if data.get('val'):
data['val'] = WANDB_ARTIFACT_PREFIX + str(Path(project) / 'val')
path = Path(data_file)
# create a _wandb.yaml file with artifacts links if both train and test set are logged
if not log_val_only:
path = (path.stem if overwrite_config else path.stem + '_wandb') + '.yaml' # updated data.yaml path
path = ROOT / 'data' / path
data.pop('download', None)
data.pop('path', None)
with open(path, 'w') as f:
yaml.safe_dump(data, f)
LOGGER.info(f"Created dataset config file {path}")
if self.job_type == 'Training': # builds correct artifact pipeline graph
if not log_val_only:
self.wandb_run.log_artifact(
self.train_artifact) # calling use_artifact downloads the dataset. NOT NEEDED!
self.wandb_run.use_artifact(self.val_artifact)
self.val_artifact.wait()
self.val_table = self.val_artifact.get('val')
self.map_val_table_path()
else:
self.wandb_run.log_artifact(self.train_artifact)
self.wandb_run.log_artifact(self.val_artifact)
return path
def map_val_table_path(self):
"""
Map the validation dataset Table like name of file -> it's id in the W&B Table.
Useful for - referencing artifacts for evaluation.
"""
self.val_table_path_map = {}
LOGGER.info("Mapping dataset")
for i, data in enumerate(tqdm(self.val_table.data)):
self.val_table_path_map[data[3]] = data[0]
def create_dataset_table(self, dataset: LoadImagesAndLabels, class_to_id: Dict[int, str], name: str = 'dataset'):
"""
Create and return W&B artifact containing W&B Table of the dataset.
arguments:
dataset -- instance of LoadImagesAndLabels class used to iterate over the data to build Table
class_to_id -- hash map that maps class ids to labels
name -- name of the artifact
returns:
dataset artifact to be logged or used
"""
# TODO: Explore multiprocessing to slpit this loop parallely| This is essential for speeding up the the logging
artifact = wandb.Artifact(name=name, type="dataset")
img_files = tqdm([dataset.path]) if isinstance(dataset.path, str) and Path(dataset.path).is_dir() else None
img_files = tqdm(dataset.im_files) if not img_files else img_files
for img_file in img_files:
if Path(img_file).is_dir():
artifact.add_dir(img_file, name='data/images')
labels_path = 'labels'.join(dataset.path.rsplit('images', 1))
artifact.add_dir(labels_path, name='data/labels')
else:
artifact.add_file(img_file, name='data/images/' + Path(img_file).name)
label_file = Path(img2label_paths([img_file])[0])
artifact.add_file(str(label_file), name='data/labels/' +
label_file.name) if label_file.exists() else None
table = wandb.Table(columns=["id", "train_image", "Classes", "name"])
class_set = wandb.Classes([{'id': id, 'name': name} for id, name in class_to_id.items()])
for si, (img, labels, paths, shapes) in enumerate(tqdm(dataset)):
box_data, img_classes = [], {}
for cls, *xywh in labels[:, 1:].tolist():
cls = int(cls)
box_data.append({
"position": {
"middle": [xywh[0], xywh[1]],
"width": xywh[2],
"height": xywh[3]},
"class_id": cls,
"box_caption": "%s" % (class_to_id[cls])})
img_classes[cls] = class_to_id[cls]
boxes = {"ground_truth": {"box_data": box_data, "class_labels": class_to_id}} # inference-space
table.add_data(si, wandb.Image(paths, classes=class_set, boxes=boxes), list(img_classes.values()),
Path(paths).name)
artifact.add(table, name)
return artifact
def log_training_progress(self, predn, path, names):
"""
Build evaluation Table. Uses reference from validation dataset table.
arguments:
predn (list): list of predictions in the native space in the format - [xmin, ymin, xmax, ymax, confidence, class]
path (str): local path of the current evaluation image
names (dict(int, str)): hash map that maps class ids to labels
"""
class_set = wandb.Classes([{'id': id, 'name': name} for id, name in names.items()])
box_data = []
avg_conf_per_class = [0] * len(self.data_dict['names'])
pred_class_count = {}
for *xyxy, conf, cls in predn.tolist():
if conf >= 0.25:
cls = int(cls)
box_data.append({
"position": {
"minX": xyxy[0],
"minY": xyxy[1],
"maxX": xyxy[2],
"maxY": xyxy[3]},
"class_id": cls,
"box_caption": f"{names[cls]} {conf:.3f}",
"scores": {
"class_score": conf},
"domain": "pixel"})
avg_conf_per_class[cls] += conf
if cls in pred_class_count:
pred_class_count[cls] += 1
else:
pred_class_count[cls] = 1
for pred_class in pred_class_count.keys():
avg_conf_per_class[pred_class] = avg_conf_per_class[pred_class] / pred_class_count[pred_class]
boxes = {"predictions": {"box_data": box_data, "class_labels": names}} # inference-space
id = self.val_table_path_map[Path(path).name]
self.result_table.add_data(self.current_epoch, id, self.val_table.data[id][1],
wandb.Image(self.val_table.data[id][1], boxes=boxes, classes=class_set),
*avg_conf_per_class)
def val_one_image(self, pred, predn, path, names, im):
"""
Log validation data for one image. updates the result Table if validation dataset is uploaded and log bbox media panel
arguments:
pred (list): list of scaled predictions in the format - [xmin, ymin, xmax, ymax, confidence, class]
predn (list): list of predictions in the native space - [xmin, ymin, xmax, ymax, confidence, class]
path (str): local path of the current evaluation image
"""
if self.val_table and self.result_table: # Log Table if Val dataset is uploaded as artifact
self.log_training_progress(predn, path, names)
if len(self.bbox_media_panel_images) < self.max_imgs_to_log and self.current_epoch > 0:
if self.current_epoch % self.bbox_interval == 0:
box_data = [{
"position": {
"minX": xyxy[0],
"minY": xyxy[1],
"maxX": xyxy[2],
"maxY": xyxy[3]},
"class_id": int(cls),
"box_caption": f"{names[int(cls)]} {conf:.3f}",
"scores": {
"class_score": conf},
"domain": "pixel"} for *xyxy, conf, cls in pred.tolist()]
boxes = {"predictions": {"box_data": box_data, "class_labels": names}} # inference-space
self.bbox_media_panel_images.append(wandb.Image(im, boxes=boxes, caption=path.name))
def log(self, log_dict):
"""
save the metrics to the logging dictionary
arguments:
log_dict (Dict) -- metrics/media to be logged in current step
"""
if self.wandb_run:
for key, value in log_dict.items():
self.log_dict[key] = value
def end_epoch(self, best_result=False):
"""
commit the log_dict, model artifacts and Tables to W&B and flush the log_dict.
arguments:
best_result (boolean): Boolean representing if the result of this evaluation is best or not
"""
if self.wandb_run:
with all_logging_disabled():
if self.bbox_media_panel_images:
self.log_dict["BoundingBoxDebugger"] = self.bbox_media_panel_images
try:
wandb.log(self.log_dict)
except BaseException as e:
LOGGER.info(
f"An error occurred in wandb logger. The training will proceed without interruption. More info\n{e}"
)
self.wandb_run.finish()
self.wandb_run = None
self.log_dict = {}
self.bbox_media_panel_images = []
if self.result_artifact:
self.result_artifact.add(self.result_table, 'result')
wandb.log_artifact(self.result_artifact,
aliases=[
'latest', 'last', 'epoch ' + str(self.current_epoch),
('best' if best_result else '')])
wandb.log({"evaluation": self.result_table})
columns = ["epoch", "id", "ground truth", "prediction"]
columns.extend(self.data_dict['names'])
self.result_table = wandb.Table(columns)
self.result_artifact = wandb.Artifact("run_" + wandb.run.id + "_progress", "evaluation")
def finish_run(self):
"""
Log metrics if any and finish the current W&B run
"""
if self.wandb_run:
if self.log_dict:
with all_logging_disabled():
wandb.log(self.log_dict)
wandb.run.finish()
@contextmanager
def all_logging_disabled(highest_level=logging.CRITICAL):
""" source - https://gist.github.com/simon-weber/7853144
A context manager that will prevent any logging messages triggered during the body from being processed.
:param highest_level: the maximum logging level in use.
This would only need to be changed if a custom level greater than CRITICAL is defined.
"""
previous_level = logging.root.manager.disable
logging.disable(highest_level)
try:
yield
finally:
logging.disable(previous_level)

238
app/yolov5/utils/loss.py Normal file
View File

@ -0,0 +1,238 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Loss functions
"""
import torch
import torch.nn as nn
from app.yolov5.utils.metrics import bbox_iou
from app.yolov5.utils.torch_utils import de_parallel
def smooth_BCE(eps=0.1): # https://github.com/ultralytics/yolov3/issues/238#issuecomment-598028441
# return positive, negative label smoothing BCE targets
return 1.0 - 0.5 * eps, 0.5 * eps
class BCEBlurWithLogitsLoss(nn.Module):
# BCEwithLogitLoss() with reduced missing label effects.
def __init__(self, alpha=0.05):
super().__init__()
self.loss_fcn = nn.BCEWithLogitsLoss(reduction='none') # must be nn.BCEWithLogitsLoss()
self.alpha = alpha
def forward(self, pred, true):
loss = self.loss_fcn(pred, true)
pred = torch.sigmoid(pred) # prob from logits
dx = pred - true # reduce only missing label effects
# dx = (pred - true).abs() # reduce missing label and false label effects
alpha_factor = 1 - torch.exp((dx - 1) / (self.alpha + 1e-4))
loss *= alpha_factor
return loss.mean()
class FocalLoss(nn.Module):
# Wraps focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)
def __init__(self, loss_fcn, gamma=1.5, alpha=0.25):
super().__init__()
self.loss_fcn = loss_fcn # must be nn.BCEWithLogitsLoss()
self.gamma = gamma
self.alpha = alpha
self.reduction = loss_fcn.reduction
self.loss_fcn.reduction = 'none' # required to apply FL to each element
def forward(self, pred, true):
loss = self.loss_fcn(pred, true)
# p_t = torch.exp(-loss)
# loss *= self.alpha * (1.000001 - p_t) ** self.gamma # non-zero power for gradient stability
# TF implementation https://github.com/tensorflow/addons/blob/v0.7.1/tensorflow_addons/losses/focal_loss.py
pred_prob = torch.sigmoid(pred) # prob from logits
p_t = true * pred_prob + (1 - true) * (1 - pred_prob)
alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha)
modulating_factor = (1.0 - p_t) ** self.gamma
loss *= alpha_factor * modulating_factor
if self.reduction == 'mean':
return loss.mean()
elif self.reduction == 'sum':
return loss.sum()
else: # 'none'
return loss
class QFocalLoss(nn.Module):
# Wraps Quality focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)
def __init__(self, loss_fcn, gamma=1.5, alpha=0.25):
super().__init__()
self.loss_fcn = loss_fcn # must be nn.BCEWithLogitsLoss()
self.gamma = gamma
self.alpha = alpha
self.reduction = loss_fcn.reduction
self.loss_fcn.reduction = 'none' # required to apply FL to each element
def forward(self, pred, true):
loss = self.loss_fcn(pred, true)
pred_prob = torch.sigmoid(pred) # prob from logits
alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha)
modulating_factor = torch.abs(true - pred_prob) ** self.gamma
loss *= alpha_factor * modulating_factor
if self.reduction == 'mean':
return loss.mean()
elif self.reduction == 'sum':
return loss.sum()
else: # 'none'
return loss
class ComputeLoss:
sort_obj_iou = False
# Compute losses
def __init__(self, model, autobalance=False):
device = next(model.parameters()).device # get model device
h = model.hyp # hyperparameters
# Define criteria
BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['cls_pw']], device=device))
BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device))
# Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3
self.cp, self.cn = smooth_BCE(eps=h.get('label_smoothing', 0.0)) # positive, negative BCE targets
# Focal loss
g = h['fl_gamma'] # focal loss gamma
if g > 0:
BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g)
m = de_parallel(model).model[-1] # Detect() module
self.balance = {3: [4.0, 1.0, 0.4]}.get(m.nl, [4.0, 1.0, 0.25, 0.06, 0.02]) # P3-P7
self.ssi = list(m.stride).index(16) if autobalance else 0 # stride 16 index
self.BCEcls, self.BCEobj, self.gr, self.hyp, self.autobalance = BCEcls, BCEobj, 1.0, h, autobalance
self.na = m.na # number of anchors
self.nc = m.nc # number of classes
self.nl = m.nl # number of layers
self.anchors = m.anchors
self.device = device
def __call__(self, p, targets): # predictions, targets
lcls = torch.zeros(1, device=self.device) # class loss
lbox = torch.zeros(1, device=self.device) # box loss
lobj = torch.zeros(1, device=self.device) # object loss
tcls, tbox, indices, anchors = self.build_targets(p, targets) # targets
# Losses
for i, pi in enumerate(p): # layer index, layer predictions
b, a, gj, gi = indices[i] # image, anchor, gridy, gridx
tobj = torch.zeros(pi.shape[:4], dtype=pi.dtype, device=self.device) # target obj
n = b.shape[0] # number of targets
# 增加部分：判断tcls里面的标签是否超界，超界赋0
for url in range(len(tcls[i])):
if tcls[i][url] > (self.nc - 1):
tcls[i][url] = 0
if n:
# pxy, pwh, _, pcls = pi[b, a, gj, gi].tensor_split((2, 4, 5), dim=1) # faster, requires torch 1.8.0
pxy, pwh, _, pcls = pi[b, a, gj, gi].split((2, 2, 1, self.nc), 1) # target-subset of predictions
# Regression
pxy = pxy.sigmoid() * 2 - 0.5
pwh = (pwh.sigmoid() * 2) ** 2 * anchors[i]
pbox = torch.cat((pxy, pwh), 1) # predicted box
iou = bbox_iou(pbox, tbox[i], CIoU=True).squeeze() # iou(prediction, target)
lbox += (1.0 - iou).mean() # iou loss
# Objectness
iou = iou.detach().clamp(0).type(tobj.dtype)
if self.sort_obj_iou:
j = iou.argsort()
b, a, gj, gi, iou = b[j], a[j], gj[j], gi[j], iou[j]
if self.gr < 1:
iou = (1.0 - self.gr) + self.gr * iou
tobj[b, a, gj, gi] = iou # iou ratio
# Classification
if self.nc > 1: # cls loss (only if multiple classes)
t = torch.full_like(pcls, self.cn, device=self.device) # targets
t[range(n), tcls[i]] = self.cp
lcls += self.BCEcls(pcls, t) # BCE
# Append targets to text file
# with open('targets.txt', 'a') as file:
# [file.write('%11.5g ' * 4 % tuple(x) + '\n') for x in torch.cat((txy[i], twh[i]), 1)]
obji = self.BCEobj(pi[..., 4], tobj)
lobj += obji * self.balance[i] # obj loss
if self.autobalance:
self.balance[i] = self.balance[i] * 0.9999 + 0.0001 / obji.detach().item()
if self.autobalance:
self.balance = [x / self.balance[self.ssi] for x in self.balance]
lbox *= self.hyp['box']
lobj *= self.hyp['obj']
lcls *= self.hyp['cls']
bs = tobj.shape[0] # batch size
return (lbox + lobj + lcls) * bs, torch.cat((lbox, lobj, lcls)).detach()
def build_targets(self, p, targets):
# Build targets for compute_loss(), input targets(image,class,x,y,w,h)
na, nt = self.na, targets.shape[0] # number of anchors, targets
tcls, tbox, indices, anch = [], [], [], []
gain = torch.ones(7, device=self.device) # normalized to gridspace gain
ai = torch.arange(na, device=self.device).float().view(na, 1).repeat(1, nt) # same as .repeat_interleave(nt)
targets = torch.cat((targets.repeat(na, 1, 1), ai[..., None]), 2) # append anchor indices
g = 0.5 # bias
off = torch.tensor(
[
[0, 0],
[1, 0],
[0, 1],
[-1, 0],
[0, -1], # j,k,l,m
# [1, 1], [1, -1], [-1, 1], [-1, -1], # jk,jm,lk,lm
],
device=self.device).float() * g # offsets
for i in range(self.nl):
anchors, shape = self.anchors[i], p[i].shape
gain[2:6] = torch.tensor(shape)[[3, 2, 3, 2]] # xyxy gain
# Match targets to anchors
t = targets * gain # shape(3,n,7)
if nt:
# Matches
r = t[..., 4:6] / anchors[:, None] # wh ratio
j = torch.max(r, 1 / r).max(2)[0] < self.hyp['anchor_t'] # compare
# j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t'] # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2))
t = t[j] # filter
# Offsets
gxy = t[:, 2:4] # grid xy
gxi = gain[[2, 3]] - gxy # inverse
j, k = ((gxy % 1 < g) & (gxy > 1)).T
l, m = ((gxi % 1 < g) & (gxi > 1)).T
j = torch.stack((torch.ones_like(j), j, k, l, m))
t = t.repeat((5, 1, 1))[j]
offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j]
else:
t = targets[0]
offsets = 0
# Define
bc, gxy, gwh, a = t.chunk(4, 1) # (image, class), grid xy, grid wh, anchors
a, (b, c) = a.long().view(-1), bc.long().T # anchors, image, class
gij = (gxy - offsets).long()
gi, gj = gij.T # grid indices
# Append
indices.append((b, a, gj.clamp_(0, shape[2] - 1), gi.clamp_(0, shape[3] - 1))) # image, anchor, grid
tbox.append(torch.cat((gxy - gij, gwh), 1)) # box
anch.append(anchors[a]) # anchors
tcls.append(c) # class
return tcls, tbox, indices, anch

367
app/yolov5/utils/metrics.py Normal file
View File

@ -0,0 +1,367 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Model validation metrics
"""
import math
import warnings
from pathlib import Path
import matplotlib.pyplot as plt
import numpy as np
import torch
from app.yolov5.utils import TryExcept, threaded
def fitness(x):
# Model fitness as a weighted combination of metrics
w = [0.0, 0.0, 0.1, 0.9] # weights for [P, R, mAP@0.5, mAP@0.5:0.95]
return (x[:, :4] * w).sum(1)
def smooth(y, f=0.05):
# Box filter of fraction f
nf = round(len(y) * f * 2) // 2 + 1 # number of filter elements (must be odd)
p = np.ones(nf // 2) # ones padding
yp = np.concatenate((p * y[0], y, p * y[-1]), 0) # y padded
return np.convolve(yp, np.ones(nf) / nf, mode='valid') # y-smoothed
def ap_per_class(tp, conf, pred_cls, target_cls, plot=False, save_dir='.', names=(), eps=1e-16):
""" Compute the average precision, given the recall and precision curves.
Source: https://github.com/rafaelpadilla/Object-Detection-Metrics.
# Arguments
tp: True positives (nparray, nx1 or nx10).
conf: Objectness value from 0-1 (nparray).
pred_cls: Predicted object classes (nparray).
target_cls: True object classes (nparray).
plot: Plot precision-recall curve at mAP@0.5
save_dir: Plot save directory
# Returns
The average precision as computed in py-faster-rcnn.
"""
# Sort by objectness
i = np.argsort(-conf)
tp, conf, pred_cls = tp[i], conf[i], pred_cls[i]
# Find unique classes
unique_classes, nt = np.unique(target_cls, return_counts=True)
nc = unique_classes.shape[0] # number of classes, number of detections
# Create Precision-Recall curve and compute AP for each class
px, py = np.linspace(0, 1, 1000), [] # for plotting
ap, p, r = np.zeros((nc, tp.shape[1])), np.zeros((nc, 1000)), np.zeros((nc, 1000))
for ci, c in enumerate(unique_classes):
i = pred_cls == c
n_l = nt[ci] # number of labels
n_p = i.sum() # number of predictions
if n_p == 0 or n_l == 0:
continue
# Accumulate FPs and TPs
fpc = (1 - tp[i]).cumsum(0)
tpc = tp[i].cumsum(0)
# Recall
recall = tpc / (n_l + eps) # recall curve
r[ci] = np.interp(-px, -conf[i], recall[:, 0], left=0) # negative x, xp because xp decreases
# Precision
precision = tpc / (tpc + fpc) # precision curve
p[ci] = np.interp(-px, -conf[i], precision[:, 0], left=1) # p at pr_score
# AP from recall-precision curve
for j in range(tp.shape[1]):
ap[ci, j], mpre, mrec = compute_ap(recall[:, j], precision[:, j])
if plot and j == 0:
py.append(np.interp(px, mrec, mpre)) # precision at mAP@0.5
# Compute F1 (harmonic mean of precision and recall)
f1 = 2 * p * r / (p + r + eps)
names = [v for k, v in names.items() if k in unique_classes] # list: only classes that have data
names = dict(enumerate(names)) # to dict
# if plot:
# plot_pr_curve(px, py, ap, Path(save_dir) / 'PR_curve.png', names)
# plot_mc_curve(px, f1, Path(save_dir) / 'F1_curve.png', names, ylabel='F1')
# plot_mc_curve(px, p, Path(save_dir) / 'P_curve.png', names, ylabel='Precision')
# plot_mc_curve(px, r, Path(save_dir) / 'R_curve.png', names, ylabel='Recall')
i = smooth(f1.mean(0), 0.1).argmax() # max F1 index
p, r, f1 = p[:, i], r[:, i], f1[:, i]
tp = (r * nt).round() # true positives
fp = (tp / (p + eps) - tp).round() # false positives
return tp, fp, p, r, f1, ap, unique_classes.astype(int)
def compute_ap(recall, precision):
""" Compute the average precision, given the recall and precision curves
# Arguments
recall: The recall curve (list)
precision: The precision curve (list)
# Returns
Average precision, precision curve, recall curve
"""
# Append sentinel values to beginning and end
mrec = np.concatenate(([0.0], recall, [1.0]))
mpre = np.concatenate(([1.0], precision, [0.0]))
# Compute the precision envelope
mpre = np.flip(np.maximum.accumulate(np.flip(mpre)))
# Integrate area under curve
method = 'interp' # methods: 'continuous', 'interp'
if method == 'interp':
x = np.linspace(0, 1, 101) # 101-point interp (COCO)
ap = np.trapz(np.interp(x, mrec, mpre), x) # integrate
else: # 'continuous'
i = np.where(mrec[1:] != mrec[:-1])[0] # points where x axis (recall) changes
ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) # area under curve
return ap, mpre, mrec
class ConfusionMatrix:
# Updated version of https://github.com/kaanakan/object_detection_confusion_matrix
def __init__(self, nc, conf=0.25, iou_thres=0.45):
self.matrix = np.zeros((nc + 1, nc + 1))
self.nc = nc # number of classes
self.conf = conf
self.iou_thres = iou_thres
def process_batch(self, detections, labels):
"""
Return intersection-over-union (Jaccard index) of boxes.
Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
Arguments:
detections (Array[N, 6]), x1, y1, x2, y2, conf, class
labels (Array[M, 5]), class, x1, y1, x2, y2
Returns:
None, updates confusion matrix accordingly
"""
if detections is None:
gt_classes = labels.int()
for gc in gt_classes:
self.matrix[self.nc, gc] += 1 # background FN
return
detections = detections[detections[:, 4] > self.conf]
gt_classes = labels[:, 0].int()
detection_classes = detections[:, 5].int()
iou = box_iou(labels[:, 1:], detections[:, :4])
x = torch.where(iou > self.iou_thres)
if x[0].shape[0]:
matches = torch.cat((torch.stack(x, 1), iou[x[0], x[1]][:, None]), 1).cpu().numpy()
if x[0].shape[0] > 1:
matches = matches[matches[:, 2].argsort()[::-1]]
matches = matches[np.unique(matches[:, 1], return_index=True)[1]]
matches = matches[matches[:, 2].argsort()[::-1]]
matches = matches[np.unique(matches[:, 0], return_index=True)[1]]
else:
matches = np.zeros((0, 3))
n = matches.shape[0] > 0
m0, m1, _ = matches.transpose().astype(int)
for i, gc in enumerate(gt_classes):
j = m0 == i
if n and sum(j) == 1:
self.matrix[detection_classes[m1[j]], gc] += 1 # correct
else:
self.matrix[self.nc, gc] += 1 # background FP
if n:
for i, dc in enumerate(detection_classes):
if not any(m1 == i):
self.matrix[dc, self.nc] += 1 # background FN
def matrix(self):
return self.matrix
def tp_fp(self):
tp = self.matrix.diagonal() # true positives
fp = self.matrix.sum(1) - tp # false positives
# fn = self.matrix.sum(0) - tp # false negatives (missed detections)
return tp[:-1], fp[:-1] # remove background class
@TryExcept('WARNING: ConfusionMatrix plot failure: ')
def plot(self, normalize=True, save_dir='', names=()):
import seaborn as sn
array = self.matrix / ((self.matrix.sum(0).reshape(1, -1) + 1E-9) if normalize else 1) # normalize columns
array[array < 0.005] = np.nan # don't annotate (would appear as 0.00)
fig, ax = plt.subplots(1, 1, figsize=(12, 9), tight_layout=True)
nc, nn = self.nc, len(names) # number of classes, names
sn.set(font_scale=1.0 if nc < 50 else 0.8) # for label size
labels = (0 < nn < 99) and (nn == nc) # apply names to ticklabels
with warnings.catch_warnings():
warnings.simplefilter('ignore') # suppress empty matrix RuntimeWarning: All-NaN slice encountered
sn.heatmap(array,
ax=ax,
annot=nc < 30,
annot_kws={
"size": 8},
cmap='Blues',
fmt='.2f',
square=True,
vmin=0.0,
xticklabels=names + ['background FP'] if labels else "auto",
yticklabels=names + ['background FN'] if labels else "auto").set_facecolor((1, 1, 1))
ax.set_ylabel('True')
ax.set_ylabel('Predicted')
ax.set_title('Confusion Matrix')
#fig.savefig(Path(save_dir) / 'confusion_matrix.png', dpi=250)
plt.close(fig)
def print(self):
for i in range(self.nc + 1):
print(' '.join(map(str, self.matrix[i])))
def bbox_iou(box1, box2, xywh=True, GIoU=False, DIoU=False, CIoU=False, eps=1e-7):
# Returns Intersection over Union (IoU) of box1(1,4) to box2(n,4)
# Get the coordinates of bounding boxes
if xywh: # transform from xywh to xyxy
(x1, y1, w1, h1), (x2, y2, w2, h2) = box1.chunk(4, 1), box2.chunk(4, 1)
w1_, h1_, w2_, h2_ = w1 / 2, h1 / 2, w2 / 2, h2 / 2
b1_x1, b1_x2, b1_y1, b1_y2 = x1 - w1_, x1 + w1_, y1 - h1_, y1 + h1_
b2_x1, b2_x2, b2_y1, b2_y2 = x2 - w2_, x2 + w2_, y2 - h2_, y2 + h2_
else: # x1, y1, x2, y2 = box1
b1_x1, b1_y1, b1_x2, b1_y2 = box1.chunk(4, 1)
b2_x1, b2_y1, b2_x2, b2_y2 = box2.chunk(4, 1)
w1, h1 = b1_x2 - b1_x1, b1_y2 - b1_y1
w2, h2 = b2_x2 - b2_x1, b2_y2 - b2_y1
# Intersection area
inter = (torch.min(b1_x2, b2_x2) - torch.max(b1_x1, b2_x1)).clamp(0) * \
(torch.min(b1_y2, b2_y2) - torch.max(b1_y1, b2_y1)).clamp(0)
# Union Area
union = w1 * h1 + w2 * h2 - inter + eps
# IoU
iou = inter / union
if CIoU or DIoU or GIoU:
cw = torch.max(b1_x2, b2_x2) - torch.min(b1_x1, b2_x1) # convex (smallest enclosing box) width
ch = torch.max(b1_y2, b2_y2) - torch.min(b1_y1, b2_y1) # convex height
if CIoU or DIoU: # Distance or Complete IoU https://arxiv.org/abs/1911.08287v1
c2 = cw ** 2 + ch ** 2 + eps # convex diagonal squared
rho2 = ((b2_x1 + b2_x2 - b1_x1 - b1_x2) ** 2 + (b2_y1 + b2_y2 - b1_y1 - b1_y2) ** 2) / 4 # center dist ** 2
if CIoU: # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47
v = (4 / math.pi ** 2) * torch.pow(torch.atan(w2 / (h2 + eps)) - torch.atan(w1 / (h1 + eps)), 2)
with torch.no_grad():
alpha = v / (v - iou + (1 + eps))
return iou - (rho2 / c2 + v * alpha) # CIoU
return iou - rho2 / c2 # DIoU
c_area = cw * ch + eps # convex area
return iou - (c_area - union) / c_area # GIoU https://arxiv.org/pdf/1902.09630.pdf
return iou # IoU
def box_area(box):
# box = xyxy(4,n)
return (box[2] - box[0]) * (box[3] - box[1])
def box_iou(box1, box2, eps=1e-7):
# https://github.com/pytorch/vision/blob/master/torchvision/ops/boxes.py
"""
Return intersection-over-union (Jaccard index) of boxes.
Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
Arguments:
box1 (Tensor[N, 4])
box2 (Tensor[M, 4])
Returns:
iou (Tensor[N, M]): the NxM matrix containing the pairwise
IoU values for every element in boxes1 and boxes2
"""
# inter(N,M) = (rb(N,M,2) - lt(N,M,2)).clamp(0).prod(2)
(a1, a2), (b1, b2) = box1[:, None].chunk(2, 2), box2.chunk(2, 1)
inter = (torch.min(a2, b2) - torch.max(a1, b1)).clamp(0).prod(2)
# IoU = inter / (area1 + area2 - inter)
return inter / (box_area(box1.T)[:, None] + box_area(box2.T) - inter + eps)
def bbox_ioa(box1, box2, eps=1e-7):
""" Returns the intersection over box2 area given box1, box2. Boxes are x1y1x2y2
box1: np.array of shape(4)
box2: np.array of shape(nx4)
returns: np.array of shape(n)
"""
# Get the coordinates of bounding boxes
b1_x1, b1_y1, b1_x2, b1_y2 = box1
b2_x1, b2_y1, b2_x2, b2_y2 = box2.T
# Intersection area
inter_area = (np.minimum(b1_x2, b2_x2) - np.maximum(b1_x1, b2_x1)).clip(0) * \
(np.minimum(b1_y2, b2_y2) - np.maximum(b1_y1, b2_y1)).clip(0)
# box2 area
box2_area = (b2_x2 - b2_x1) * (b2_y2 - b2_y1) + eps
# Intersection over box2 area
return inter_area / box2_area
def wh_iou(wh1, wh2, eps=1e-7):
# Returns the nxm IoU matrix. wh1 is nx2, wh2 is mx2
wh1 = wh1[:, None] # [N,1,2]
wh2 = wh2[None] # [1,M,2]
inter = torch.min(wh1, wh2).prod(2) # [N,M]
return inter / (wh1.prod(2) + wh2.prod(2) - inter + eps) # iou = inter / (area1 + area2 - inter)
# Plots ----------------------------------------------------------------------------------------------------------------
@threaded
def plot_pr_curve(px, py, ap, save_dir=Path('pr_curve.png'), names=()):
# Precision-recall curve
fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True)
py = np.stack(py, axis=1)
if 0 < len(names) < 21: # display per-class legend if < 21 classes
for i, y in enumerate(py.T):
ax.plot(px, y, linewidth=1, label=f'{names[i]} {ap[i, 0]:.3f}') # plot(recall, precision)
else:
ax.plot(px, py, linewidth=1, color='grey') # plot(recall, precision)
ax.plot(px, py.mean(1), linewidth=3, color='blue', label='all classes %.3f mAP@0.5' % ap[:, 0].mean())
ax.set_xlabel('Recall')
ax.set_ylabel('Precision')
ax.set_xlim(0, 1)
ax.set_ylim(0, 1)
ax.legend(bbox_to_anchor=(1.04, 1), loc="upper left")
ax.set_title('Precision-Recall Curve')
fig.savefig(save_dir, dpi=250)
plt.close(fig)
@threaded
def plot_mc_curve(px, py, save_dir=Path('mc_curve.png'), names=(), xlabel='Confidence', ylabel='Metric'):
# Metric-confidence curve
fig, ax = plt.subplots(1, 1, figsize=(9, 6), tight_layout=True)
if 0 < len(names) < 21: # display per-class legend if < 21 classes
for i, y in enumerate(py):
ax.plot(px, y, linewidth=1, label=f'{names[i]}') # plot(confidence, metric)
else:
ax.plot(px, py.T, linewidth=1, color='grey') # plot(confidence, metric)
y = smooth(py.mean(0), 0.05)
ax.plot(px, y, linewidth=3, color='blue', label=f'all classes {y.max():.2f} at {px[y.argmax()]:.3f}')
ax.set_xlabel(xlabel)
ax.set_ylabel(ylabel)
ax.set_xlim(0, 1)
ax.set_ylim(0, 1)
ax.legend(bbox_to_anchor=(1.04, 1), loc="upper left")
ax.set_title(f'{ylabel}-Confidence Curve')
fig.savefig(save_dir, dpi=250)
plt.close(fig)

519
app/yolov5/utils/plots.py Normal file
View File

@ -0,0 +1,519 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
Plotting utils
"""
import contextlib
import math
import os
from copy import copy
from pathlib import Path
from urllib.error import URLError
import cv2
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sn
import torch
from PIL import Image, ImageDraw, ImageFont
from app.yolov5.utils import TryExcept, threaded
from app.yolov5.utils.general import (CONFIG_DIR, FONT, LOGGER, check_font, check_requirements, clip_coords, increment_path,
is_ascii, xywh2xyxy, xyxy2xywh)
from app.yolov5.utils.metrics import fitness
# Settings
RANK = int(os.getenv('RANK', -1))
matplotlib.rc('font', **{'size': 11})
matplotlib.use('Agg') # for writing to files only
class Colors:
# Ultralytics color palette https://ultralytics.com/
def __init__(self):
# hex = matplotlib.colors.TABLEAU_COLORS.values()
hexs = ('FF3838', 'FF9D97', 'FF701F', 'FFB21D', 'CFD231', '48F90A', '92CC17', '3DDB86', '1A9334', '00D4BB',
'2C99A8', '00C2FF', '344593', '6473FF', '0018EC', '8438FF', '520085', 'CB38FF', 'FF95C8', 'FF37C7')
self.palette = [self.hex2rgb(f'#{c}') for c in hexs]
self.n = len(self.palette)
def __call__(self, i, bgr=False):
c = self.palette[int(i) % self.n]
return (c[2], c[1], c[0]) if bgr else c
@staticmethod
def hex2rgb(h): # rgb order (PIL)
return tuple(int(h[1 + i:1 + i + 2], 16) for i in (0, 2, 4))
colors = Colors() # create instance for 'from utils.plots import colors'
def check_pil_font(font=FONT, size=10):
# Return a PIL TrueType Font, downloading to CONFIG_DIR if necessary
font = Path(font)
font = font if font.exists() else (CONFIG_DIR / font.name)
try:
return ImageFont.truetype(str(font) if font.exists() else font.name, size)
except Exception: # download if missing
try:
check_font(font)
return ImageFont.truetype(str(font), size)
except TypeError:
check_requirements('Pillow>=8.4.0') # known issue https://github.com/ultralytics/yolov5/issues/5374
except URLError: # not online
return ImageFont.load_default()
class Annotator:
# YOLOv5 Annotator for train/val mosaics and jpgs and detect/hub inference annotations
def __init__(self, im, line_width=None, font_size=None, font='Arial.ttf', pil=False, example='abc'):
assert im.data.contiguous, 'Image not contiguous. Apply np.ascontiguousarray(im) to Annotator() input images.'
non_ascii = not is_ascii(example) # non-latin labels, i.e. asian, arabic, cyrillic
self.pil = pil or non_ascii
if self.pil: # use PIL
self.im = im if isinstance(im, Image.Image) else Image.fromarray(im)
self.draw = ImageDraw.Draw(self.im)
self.font = check_pil_font(font='Arial.Unicode.ttf' if non_ascii else font,
size=font_size or max(round(sum(self.im.size) / 2 * 0.035), 12))
else: # use cv2
self.im = im
self.lw = line_width or max(round(sum(im.shape) / 2 * 0.003), 2) # line width
def box_label(self, box, label='', color=(128, 128, 128), txt_color=(0, 255, 0)):
# Add one xyxy box to image with label
if self.pil or not is_ascii(label):
self.draw.rectangle(box, width=self.lw, outline=color) # box
if label:
w, h = self.font.getsize(label) # text width, height
outside = box[1] - h >= 0 # label fits outside box
self.draw.rectangle(
(box[0], box[1] - h if outside else box[1], box[0] + w + 1,
box[1] + 1 if outside else box[1] + h + 1),
fill=color,
)
# self.draw.text((box[0], box[1]), label, fill=txt_color, font=self.font, anchor='ls') # for PIL>8.0
self.draw.text((box[0], box[1] - h if outside else box[1]), label, fill=txt_color, font=self.font)
else: # cv2
p1, p2 = (int(box[0]), int(box[1])), (int(box[2]), int(box[3]))
cv2.rectangle(self.im, p1, p2, color, thickness=self.lw, lineType=cv2.LINE_AA)
if label:
tf = max(self.lw - 1, 1) # font thickness
w, h = cv2.getTextSize(label, 0, fontScale=self.lw / 3, thickness=tf)[0] # text width, height
outside = p1[1] - h >= 3
p2 = p1[0] + w, p1[1] - h - 3 if outside else p1[1] + h + 3
cv2.rectangle(self.im, p1, p2, color, -1, cv2.LINE_AA) # filled
cv2.putText(self.im,
label, (p1[0], p1[1] - 2 if outside else p1[1] + h + 2),
0,
self.lw / 3,
txt_color,
thickness=tf,
lineType=cv2.LINE_AA)
def rectangle(self, xy, fill=None, outline=None, width=1):
# Add rectangle to image (PIL-only)
self.draw.rectangle(xy, fill, outline, width)
def text(self, xy, text, txt_color=(255, 255, 255), anchor='top'):
# Add text to image (PIL-only)
if anchor == 'bottom': # start y from font bottom
w, h = self.font.getsize(text) # text width, height
xy[1] += 1 - h
self.draw.text(xy, text, fill=txt_color, font=self.font)
def result(self):
# Return annotated image as array
return np.asarray(self.im)
def feature_visualization(x, module_type, stage, n=32, save_dir=Path('runs/detect/exp')):
"""
x: Features to be visualized
module_type: Module type
stage: Module stage within model
n: Maximum number of feature maps to plot
save_dir: Directory to save results
"""
if 'Detect' not in module_type:
batch, channels, height, width = x.shape # batch, channels, height, width
if height > 1 and width > 1:
f = save_dir / f"stage{stage}_{module_type.split('.')[-1]}_features.png" # filename
blocks = torch.chunk(x[0].cpu(), channels, dim=0) # select batch index 0, block by channels
n = min(n, channels) # number of plots
fig, ax = plt.subplots(math.ceil(n / 8), 8, tight_layout=True) # 8 rows x n/8 cols
ax = ax.ravel()
plt.subplots_adjust(wspace=0.05, hspace=0.05)
for i in range(n):
ax[i].imshow(blocks[i].squeeze()) # cmap='gray'
ax[i].axis('off')
LOGGER.info(f'Saving {f}... ({n}/{channels})')
plt.title('Features')
plt.savefig(f, dpi=300, bbox_inches='tight')
plt.close()
np.save(str(f.with_suffix('.npy')), x[0].cpu().numpy()) # npy save
def hist2d(x, y, n=100):
# 2d histogram used in labels.png and evolve.png
xedges, yedges = np.linspace(x.min(), x.max(), n), np.linspace(y.min(), y.max(), n)
hist, xedges, yedges = np.histogram2d(x, y, (xedges, yedges))
xidx = np.clip(np.digitize(x, xedges) - 1, 0, hist.shape[0] - 1)
yidx = np.clip(np.digitize(y, yedges) - 1, 0, hist.shape[1] - 1)
return np.log(hist[xidx, yidx])
def butter_lowpass_filtfilt(data, cutoff=1500, fs=50000, order=5):
from scipy.signal import butter, filtfilt
# https://stackoverflow.com/questions/28536191/how-to-filter-smooth-with-scipy-numpy
def butter_lowpass(cutoff, fs, order):
nyq = 0.5 * fs
normal_cutoff = cutoff / nyq
return butter(order, normal_cutoff, btype='low', analog=False)
b, a = butter_lowpass(cutoff, fs, order=order)
return filtfilt(b, a, data) # forward-backward filter
def output_to_target(output):
# Convert model output to target format [batch_id, class_id, x, y, w, h, conf]
targets = []
for i, o in enumerate(output):
targets.extend([i, cls, *list(*xyxy2xywh(np.array(box)[None])), conf] for *box, conf, cls in o.cpu().numpy())
return np.array(targets)
@threaded
def plot_images(images, targets, paths=None, fname='images.jpg', names=None, max_size=1920, max_subplots=16):
# Plot image grid with labels
if isinstance(images, torch.Tensor):
images = images.cpu().float().numpy()
if isinstance(targets, torch.Tensor):
targets = targets.cpu().numpy()
if np.max(images[0]) <= 1:
images *= 255 # de-normalise (optional)
bs, _, h, w = images.shape # batch size, _, height, width
bs = min(bs, max_subplots) # limit plot images
ns = np.ceil(bs ** 0.5) # number of subplots (square)
# Build Image
mosaic = np.full((int(ns * h), int(ns * w), 3), 255, dtype=np.uint8) # init
for i, im in enumerate(images):
if i == max_subplots: # if last batch has fewer images than we expect
break
x, y = int(w * (i // ns)), int(h * (i % ns)) # block origin
im = im.transpose(1, 2, 0)
mosaic[y:y + h, x:x + w, :] = im
# Resize (optional)
scale = max_size / ns / max(h, w)
if scale < 1:
h = math.ceil(scale * h)
w = math.ceil(scale * w)
mosaic = cv2.resize(mosaic, tuple(int(x * ns) for x in (w, h)))
# Annotate
fs = int((h + w) * ns * 0.01) # font size
annotator = Annotator(mosaic, line_width=round(fs/5), font_size=fs*2, pil=True, example=names)
for i in range(i + 1):
x, y = int(w * (i // ns)), int(h * (i % ns)) # block origin
annotator.rectangle([x, y, x + w, y + h], None, (255, 255, 255), width=2) # borders
if paths:
annotator.text((x + 5, y + 5), text=Path(paths[i]).name[:40], txt_color=(220, 220, 220)) # filenames
if len(targets) > 0:
ti = targets[targets[:, 0] == i] # image targets
boxes = xywh2xyxy(ti[:, 2:6]).T
classes = ti[:, 1].astype('int')
labels = ti.shape[1] == 6 # labels if no conf column
conf = None if labels else ti[:, 6] # check for confidence presence (label vs pred)
if boxes.shape[1]:
if boxes.max() <= 1.01: # if normalized with tolerance 0.01
boxes[[0, 2]] *= w # scale to pixels
boxes[[1, 3]] *= h
elif scale < 1: # absolute coords need scale if image scales
boxes *= scale
boxes[[0, 2]] += x
boxes[[1, 3]] += y
for j, box in enumerate(boxes.T.tolist()):
cls = classes[j]
color = colors(cls)
cls = names[cls] if names else cls
if labels or conf[j] > 0.25: # 0.25 conf thresh
label = f'{cls}' if labels else f'{cls} {conf[j]:.1f}'
annotator.box_label(box, label, color=color)
annotator.im.save(fname) # save
def plot_lr_scheduler(optimizer, scheduler, epochs=300, save_dir=''):
# Plot LR simulating training for full epochs
optimizer, scheduler = copy(optimizer), copy(scheduler) # do not modify originals
y = []
for _ in range(epochs):
scheduler.step()
y.append(optimizer.param_groups[0]['lr'])
plt.plot(y, '.-', label='LR')
plt.xlabel('epoch')
plt.ylabel('LR')
plt.grid()
plt.xlim(0, epochs)
plt.ylim(0)
plt.savefig(Path(save_dir) / 'LR.png', dpi=200)
plt.close()
def plot_val_txt(): # from utils.plots import *; plot_val()
# Plot val.txt histograms
x = np.loadtxt('val.txt', dtype=np.float32)
box = xyxy2xywh(x[:, :4])
cx, cy = box[:, 0], box[:, 1]
fig, ax = plt.subplots(1, 1, figsize=(6, 6), tight_layout=True)
ax.hist2d(cx, cy, bins=600, cmax=10, cmin=0)
ax.set_aspect('equal')
plt.savefig('hist2d.png', dpi=300)
fig, ax = plt.subplots(1, 2, figsize=(12, 6), tight_layout=True)
ax[0].hist(cx, bins=600)
ax[1].hist(cy, bins=600)
plt.savefig('hist1d.png', dpi=200)
def plot_targets_txt(): # from utils.plots import *; plot_targets_txt()
# Plot targets.txt histograms
x = np.loadtxt('targets.txt', dtype=np.float32).T
s = ['x targets', 'y targets', 'width targets', 'height targets']
fig, ax = plt.subplots(2, 2, figsize=(8, 8), tight_layout=True)
ax = ax.ravel()
for i in range(4):
ax[i].hist(x[i], bins=100, label=f'{x[i].mean():.3g} +/- {x[i].std():.3g}')
ax[i].legend()
ax[i].set_title(s[i])
plt.savefig('targets.jpg', dpi=200)
def plot_val_study(file='', dir='', x=None): # from utils.plots import *; plot_val_study()
# Plot file=study.txt generated by val.py (or plot all study*.txt in dir)
save_dir = Path(file).parent if file else Path(dir)
plot2 = False # plot additional results
if plot2:
ax = plt.subplots(2, 4, figsize=(10, 6), tight_layout=True)[1].ravel()
fig2, ax2 = plt.subplots(1, 1, figsize=(8, 4), tight_layout=True)
# for f in [save_dir / f'study_coco_{x}.txt' for x in ['yolov5n6', 'yolov5s6', 'yolov5m6', 'yolov5l6', 'yolov5x6']]:
for f in sorted(save_dir.glob('study*.txt')):
y = np.loadtxt(f, dtype=np.float32, usecols=[0, 1, 2, 3, 7, 8, 9], ndmin=2).T
x = np.arange(y.shape[1]) if x is None else np.array(x)
if plot2:
s = ['P', 'R', 'mAP@.5', 'mAP@.5:.95', 't_preprocess (ms/img)', 't_inference (ms/img)', 't_NMS (ms/img)']
for i in range(7):
ax[i].plot(x, y[i], '.-', linewidth=2, markersize=8)
ax[i].set_title(s[i])
j = y[3].argmax() + 1
ax2.plot(y[5, 1:j],
y[3, 1:j] * 1E2,
'.-',
linewidth=2,
markersize=8,
label=f.stem.replace('study_coco_', '').replace('yolo', 'YOLO'))
ax2.plot(1E3 / np.array([209, 140, 97, 58, 35, 18]), [34.6, 40.5, 43.0, 47.5, 49.7, 51.5],
'k.-',
linewidth=2,
markersize=8,
alpha=.25,
label='EfficientDet')
ax2.grid(alpha=0.2)
ax2.set_yticks(np.arange(20, 60, 5))
ax2.set_xlim(0, 57)
ax2.set_ylim(25, 55)
ax2.set_xlabel('GPU Speed (ms/img)')
ax2.set_ylabel('COCO AP val')
ax2.legend(loc='lower right')
f = save_dir / 'study.png'
print(f'Saving {f}...')
plt.savefig(f, dpi=300)
@TryExcept() # known issue https://github.com/ultralytics/yolov5/issues/5395
def plot_labels(labels, names=(), save_dir=Path('')):
# plot dataset labels
LOGGER.info(f"Plotting labels to {save_dir / 'labels.jpg'}... ")
c, b = labels[:, 0], labels[:, 1:].transpose() # classes, boxes
nc = int(c.max() + 1) # number of classes
x = pd.DataFrame(b.transpose(), columns=['x', 'y', 'width', 'height'])
# seaborn correlogram
sn.pairplot(x, corner=True, diag_kind='auto', kind='hist', diag_kws=dict(bins=50), plot_kws=dict(pmax=0.9))
plt.savefig(save_dir / 'labels_correlogram.jpg', dpi=200)
plt.close()
# matplotlib labels
matplotlib.use('svg') # faster
ax = plt.subplots(2, 2, figsize=(8, 8), tight_layout=True)[1].ravel()
y = ax[0].hist(c, bins=np.linspace(0, nc, nc + 1) - 0.5, rwidth=0.8)
with contextlib.suppress(Exception): # color histogram bars by class
[y[2].patches[i].set_color([x / 255 for x in colors(i)]) for i in range(nc)] # known issue #3195
ax[0].set_ylabel('instances')
if 0 < len(names) < 30:
ax[0].set_xticks(range(len(names)))
ax[0].set_xticklabels(names, rotation=90, fontsize=10)
else:
ax[0].set_xlabel('classes')
sn.histplot(x, x='x', y='y', ax=ax[2], bins=50, pmax=0.9)
sn.histplot(x, x='width', y='height', ax=ax[3], bins=50, pmax=0.9)
# rectangles
labels[:, 1:3] = 0.5 # center
labels[:, 1:] = xywh2xyxy(labels[:, 1:]) * 2000
img = Image.fromarray(np.ones((2000, 2000, 3), dtype=np.uint8) * 255)
for cls, *box in labels[:1000]:
ImageDraw.Draw(img).rectangle(box, width=1, outline=colors(cls)) # plot
ax[1].imshow(img)
ax[1].axis('off')
for a in [0, 1, 2, 3]:
for s in ['top', 'right', 'left', 'bottom']:
ax[a].spines[s].set_visible(False)
plt.savefig(save_dir / 'labels.jpg', dpi=200)
matplotlib.use('Agg')
plt.close()
def imshow_cls(im, labels=None, pred=None, names=None, nmax=25, verbose=False, f=Path('images.jpg')):
# Show classification image grid with labels (optional) and predictions (optional)
from yolov5.utils.augmentations import denormalize
names = names or [f'class{i}' for i in range(1000)]
blocks = torch.chunk(denormalize(im.clone()).cpu().float(), len(im),
dim=0) # select batch index 0, block by channels
n = min(len(blocks), nmax) # number of plots
m = min(8, round(n ** 0.5)) # 8 x 8 default
fig, ax = plt.subplots(math.ceil(n / m), m) # 8 rows x n/8 cols
ax = ax.ravel() if m > 1 else [ax]
# plt.subplots_adjust(wspace=0.05, hspace=0.05)
for i in range(n):
ax[i].imshow(blocks[i].squeeze().permute((1, 2, 0)).numpy().clip(0.0, 1.0))
ax[i].axis('off')
if labels is not None:
s = names[labels[i]] + (f'—{names[pred[i]]}' if pred is not None else '')
ax[i].set_title(s, fontsize=8, verticalalignment='top')
plt.savefig(f, dpi=300, bbox_inches='tight')
plt.close()
if verbose:
LOGGER.info(f"Saving {f}")
if labels is not None:
LOGGER.info('True: ' + ' '.join(f'{names[i]:3s}' for i in labels[:nmax]))
if pred is not None:
LOGGER.info('Predicted:' + ' '.join(f'{names[i]:3s}' for i in pred[:nmax]))
return f
def plot_evolve(evolve_csv='path/to/evolve.csv'): # from utils.plots import *; plot_evolve()
# Plot evolve.csv hyp evolution results
evolve_csv = Path(evolve_csv)
data = pd.read_csv(evolve_csv)
keys = [x.strip() for x in data.columns]
x = data.values
f = fitness(x)
j = np.argmax(f) # max fitness index
plt.figure(figsize=(10, 12), tight_layout=True)
matplotlib.rc('font', **{'size': 8})
print(f'Best results from row {j} of {evolve_csv}:')
for i, k in enumerate(keys[7:]):
v = x[:, 7 + i]
mu = v[j] # best single result
plt.subplot(6, 5, i + 1)
plt.scatter(v, f, c=hist2d(v, f, 20), cmap='viridis', alpha=.8, edgecolors='none')
plt.plot(mu, f.max(), 'k+', markersize=15)
plt.title(f'{k} = {mu:.3g}', fontdict={'size': 9}) # limit to 40 characters
if i % 5 != 0:
plt.yticks([])
print(f'{k:>15}: {mu:.3g}')
f = evolve_csv.with_suffix('.png') # filename
plt.savefig(f, dpi=200)
plt.close()
print(f'Saved {f}')
def plot_results(file='path/to/results.csv', dir=''):
# Plot training results.csv. Usage: from utils.plots import *; plot_results('path/to/results.csv')
save_dir = Path(file).parent if file else Path(dir)
fig, ax = plt.subplots(2, 5, figsize=(12, 6), tight_layout=True)
ax = ax.ravel()
files = list(save_dir.glob('results*.csv'))
assert len(files), f'No results.csv files found in {save_dir.resolve()}, nothing to plot.'
for f in files:
try:
data = pd.read_csv(f)
s = [x.strip() for x in data.columns]
x = data.values[:, 0]
for i, j in enumerate([1, 2, 3, 4, 5, 8, 9, 10, 6, 7]):
y = data.values[:, j].astype('float')
# y[y == 0] = np.nan # don't show zero values
ax[i].plot(x, y, marker='.', label=f.stem, linewidth=2, markersize=8)
ax[i].set_title(s[j], fontsize=12)
# if j in [8, 9, 10]: # share train and val loss y axes
# ax[i].get_shared_y_axes().join(ax[i], ax[i - 5])
except Exception as e:
LOGGER.info(f'Warning: Plotting error for {f}: {e}')
ax[1].legend()
fig.savefig(save_dir / 'results.png', dpi=200)
plt.close()
def profile_idetection(start=0, stop=0, labels=(), save_dir=''):
# Plot iDetection '*.txt' per-image logs. from utils.plots import *; profile_idetection()
ax = plt.subplots(2, 4, figsize=(12, 6), tight_layout=True)[1].ravel()
s = ['Images', 'Free Storage (GB)', 'RAM Usage (GB)', 'Battery', 'dt_raw (ms)', 'dt_smooth (ms)', 'real-world FPS']
files = list(Path(save_dir).glob('frames*.txt'))
for fi, f in enumerate(files):
try:
results = np.loadtxt(f, ndmin=2).T[:, 90:-30] # clip first and last rows
n = results.shape[1] # number of rows
x = np.arange(start, min(stop, n) if stop else n)
results = results[:, x]
t = (results[0] - results[0].min()) # set t0=0s
results[0] = x
for i, a in enumerate(ax):
if i < len(results):
label = labels[fi] if len(labels) else f.stem.replace('frames_', '')
a.plot(t, results[i], marker='.', label=label, linewidth=1, markersize=5)
a.set_title(s[i])
a.set_xlabel('time (s)')
# if fi == len(files) - 1:
# a.set_ylim(bottom=0)
for side in ['top', 'right']:
a.spines[side].set_visible(False)
else:
a.remove()
except Exception as e:
print(f'Warning: Plotting error for {f}; {e}')
ax[1].legend()
plt.savefig(Path(save_dir) / 'idetection_profile.png', dpi=200)
def save_one_box(xyxy, im, file=Path('im.jpg'), gain=1.02, pad=10, square=False, BGR=False, save=True):
# Save image crop as {file} with crop size multiple {gain} and {pad} pixels. Save and/or return crop
xyxy = torch.tensor(xyxy).view(-1, 4)
b = xyxy2xywh(xyxy) # boxes
if square:
b[:, 2:] = b[:, 2:].max(1)[0].unsqueeze(1) # attempt rectangle to square
b[:, 2:] = b[:, 2:] * gain + pad # box wh * gain + pad
xyxy = xywh2xyxy(b).long()
clip_coords(xyxy, im.shape)
crop = im[int(xyxy[0, 1]):int(xyxy[0, 3]), int(xyxy[0, 0]):int(xyxy[0, 2]), ::(1 if BGR else -1)]
if save:
file.parent.mkdir(parents=True, exist_ok=True) # make directory
f = str(increment_path(file).with_suffix('.jpg'))
# cv2.imwrite(f, crop) # save BGR, https://github.com/ultralytics/yolov5/issues/7007 chroma subsampling issue
Image.fromarray(crop[..., ::-1]).save(f, quality=95, subsampling=0) # save RGB
return crop

430
app/yolov5/utils/torch_utils.py Normal file
View File

@ -0,0 +1,430 @@
# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
PyTorch utils
"""
import math
import os
import platform
import subprocess
import time
import warnings
from contextlib import contextmanager
from copy import deepcopy
from pathlib import Path
import torch
import torch.distributed as dist
import torch.nn as nn
import torch.nn.functional as F
from torch.nn.parallel import DistributedDataParallel as DDP
from app.yolov5.utils.general import LOGGER, check_version, colorstr, file_date, git_describe
LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1)) # https://pytorch.org/docs/stable/elastic/run.html
RANK = int(os.getenv('RANK', -1))
WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1))
try:
import thop # for FLOPs computation
except ImportError:
thop = None
# Suppress PyTorch warnings
warnings.filterwarnings('ignore', message='User provided device_type of \'cuda\', but CUDA is not available. Disabling')
def smart_inference_mode(torch_1_9=check_version(torch.__version__, '1.9.0')):
# Applies torch.inference_mode() decorator if torch>=1.9.0 else torch.no_grad() decorator
def decorate(fn):
return (torch.inference_mode if torch_1_9 else torch.no_grad)()(fn)
return decorate
def smartCrossEntropyLoss(label_smoothing=0.0):
# Returns nn.CrossEntropyLoss with label smoothing enabled for torch>=1.10.0
if check_version(torch.__version__, '1.10.0'):
return nn.CrossEntropyLoss(label_smoothing=label_smoothing)
if label_smoothing > 0:
LOGGER.warning(f'WARNING: label smoothing {label_smoothing} requires torch>=1.10.0')
return nn.CrossEntropyLoss()
def smart_DDP(model):
# Model DDP creation with checks
assert not check_version(torch.__version__, '1.12.0', pinned=True), \
'torch==1.12.0 torchvision==0.13.0 DDP training is not supported due to a known issue. ' \
'Please upgrade or downgrade torch to use DDP. See https://github.com/ultralytics/yolov5/issues/8395'
if check_version(torch.__version__, '1.11.0'):
return DDP(model, device_ids=[LOCAL_RANK], output_device=LOCAL_RANK, static_graph=True)
else:
return DDP(model, device_ids=[LOCAL_RANK], output_device=LOCAL_RANK)
def reshape_classifier_output(model, n=1000):
# Update a TorchVision classification model to class count 'n' if required
from models.common import Classify
name, m = list((model.model if hasattr(model, 'model') else model).named_children())[-1] # last module
if isinstance(m, Classify): # YOLOv5 Classify() head
if m.linear.out_features != n:
m.linear = nn.Linear(m.linear.in_features, n)
elif isinstance(m, nn.Linear): # ResNet, EfficientNet
if m.out_features != n:
setattr(model, name, nn.Linear(m.in_features, n))
elif isinstance(m, nn.Sequential):
types = [type(x) for x in m]
if nn.Linear in types:
i = types.index(nn.Linear) # nn.Linear index
if m[i].out_features != n:
m[i] = nn.Linear(m[i].in_features, n)
elif nn.Conv2d in types:
i = types.index(nn.Conv2d) # nn.Conv2d index
if m[i].out_channels != n:
m[i] = nn.Conv2d(m[i].in_channels, n, m[i].kernel_size, m[i].stride, bias=m[i].bias)
@contextmanager
def torch_distributed_zero_first(local_rank: int):
# Decorator to make all processes in distributed training wait for each local_master to do something
if local_rank not in [-1, 0]:
dist.barrier(device_ids=[local_rank])
yield
if local_rank == 0:
dist.barrier(device_ids=[0])
def device_count():
# Returns number of CUDA devices available. Safe version of torch.cuda.device_count(). Supports Linux and Windows
assert platform.system() in ('Linux', 'Windows'), 'device_count() only supported on Linux or Windows'
try:
cmd = 'nvidia-smi -L | wc -l' if platform.system() == 'Linux' else 'nvidia-smi -L | find /c /v ""' # Windows
return int(subprocess.run(cmd, shell=True, capture_output=True, check=True).stdout.decode().split()[-1])
except Exception:
return 0
def select_device(device='', batch_size=0, newline=True):
# device = None or 'cpu' or 0 or '0' or '0,1,2,3'
s = f'YOLOv5 🚀 {git_describe() or file_date()} Python-{platform.python_version()} torch-{torch.__version__} '
device = str(device).strip().lower().replace('cuda:', '').replace('none', '') # to string, 'cuda:0' to '0'
cpu = device == 'cpu'
mps = device == 'mps' # Apple Metal Performance Shaders (MPS)
if cpu or mps:
os.environ['CUDA_VISIBLE_DEVICES'] = '-1' # force torch.cuda.is_available() = False
elif device: # non-cpu device requested
os.environ['CUDA_VISIBLE_DEVICES'] = device # set environment variable - must be before assert is_available()
print(torch.cuda.is_available())
assert torch.cuda.is_available() and torch.cuda.device_count() >= len(device.replace(',', '')), \
f"Invalid CUDA '--device {device}' requested, use '--device cpu' or pass valid CUDA device(s)"
if not cpu and not mps and torch.cuda.is_available(): # prefer GPU if available
devices = device.split(',') if device else '0' # range(torch.cuda.device_count()) # i.e. 0,1,6,7
n = len(devices) # device count
if n > 1 and batch_size > 0: # check batch_size is divisible by device_count
assert batch_size % n == 0, f'batch-size {batch_size} not multiple of GPU count {n}'
space = ' ' * (len(s) + 1)
for i, d in enumerate(devices):
p = torch.cuda.get_device_properties(i)
s += f"{'' if i == 0 else space}CUDA:{d} ({p.name}, {p.total_memory / (1 << 20):.0f}MiB)\n" # bytes to MB
arg = 'cuda:0'
elif mps and getattr(torch, 'has_mps', False) and torch.backends.mps.is_available(): # prefer MPS if available
s += 'MPS\n'
arg = 'mps'
else: # revert to CPU
s += 'CPU\n'
arg = 'cpu'
if not newline:
s = s.rstrip()
LOGGER.info(s)
return torch.device(arg)
def time_sync():
# PyTorch-accurate time
if torch.cuda.is_available():
torch.cuda.synchronize()
return time.time()
def profile(input, ops, n=10, device=None):
""" YOLOv5 speed/memory/FLOPs profiler
Usage:
input = torch.randn(16, 3, 640, 640)
m1 = lambda x: x * torch.sigmoid(x)
m2 = nn.SiLU()
profile(input, [m1, m2], n=100) # profile over 100 iterations
"""
results = []
if not isinstance(device, torch.device):
device = select_device(device)
print(f"{'Params':>12s}{'GFLOPs':>12s}{'GPU_mem (GB)':>14s}{'forward (ms)':>14s}{'backward (ms)':>14s}"
f"{'input':>24s}{'output':>24s}")
for x in input if isinstance(input, list) else [input]:
x = x.to(device)
x.requires_grad = True
for m in ops if isinstance(ops, list) else [ops]:
m = m.to(device) if hasattr(m, 'to') else m # device
m = m.half() if hasattr(m, 'half') and isinstance(x, torch.Tensor) and x.dtype is torch.float16 else m
tf, tb, t = 0, 0, [0, 0, 0] # dt forward, backward
try:
flops = thop.profile(m, inputs=(x,), verbose=False)[0] / 1E9 * 2 # GFLOPs
except Exception:
flops = 0
try:
for _ in range(n):
t[0] = time_sync()
y = m(x)
t[1] = time_sync()
try:
_ = (sum(yi.sum() for yi in y) if isinstance(y, list) else y).sum().backward()
t[2] = time_sync()
except Exception: # no backward method
# print(e) # for debug
t[2] = float('nan')
tf += (t[1] - t[0]) * 1000 / n # ms per op forward
tb += (t[2] - t[1]) * 1000 / n # ms per op backward
mem = torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0 # (GB)
s_in, s_out = (tuple(x.shape) if isinstance(x, torch.Tensor) else 'list' for x in (x, y)) # shapes
p = sum(x.numel() for x in m.parameters()) if isinstance(m, nn.Module) else 0 # parameters
print(f'{p:12}{flops:12.4g}{mem:>14.3f}{tf:14.4g}{tb:14.4g}{str(s_in):>24s}{str(s_out):>24s}')
results.append([p, flops, mem, tf, tb, s_in, s_out])
except Exception as e:
print(e)
results.append(None)
torch.cuda.empty_cache()
return results
def is_parallel(model):
# Returns True if model is of type DP or DDP
return type(model) in (nn.parallel.DataParallel, nn.parallel.DistributedDataParallel)
def de_parallel(model):
# De-parallelize a model: returns single-GPU model if model is of type DP or DDP
return model.module if is_parallel(model) else model
def initialize_weights(model):
for m in model.modules():
t = type(m)
if t is nn.Conv2d:
pass # nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
elif t is nn.BatchNorm2d:
m.eps = 1e-3
m.momentum = 0.03
elif t in [nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU]:
m.inplace = True
def find_modules(model, mclass=nn.Conv2d):
# Finds layer indices matching module class 'mclass'
return [i for i, m in enumerate(model.module_list) if isinstance(m, mclass)]
def sparsity(model):
# Return global model sparsity
a, b = 0, 0
for p in model.parameters():
a += p.numel()
b += (p == 0).sum()
return b / a
def prune(model, amount=0.3):
# Prune model to requested global sparsity
import torch.nn.utils.prune as prune
for name, m in model.named_modules():
if isinstance(m, nn.Conv2d):
prune.l1_unstructured(m, name='weight', amount=amount) # prune
prune.remove(m, 'weight') # make permanent
LOGGER.info(f'Model pruned to {sparsity(model):.3g} global sparsity')
def fuse_conv_and_bn(conv, bn):
# Fuse Conv2d() and BatchNorm2d() layers https://tehnokv.com/posts/fusing-batchnorm-and-conv/
fusedconv = nn.Conv2d(conv.in_channels,
conv.out_channels,
kernel_size=conv.kernel_size,
stride=conv.stride,
padding=conv.padding,
groups=conv.groups,
bias=True).requires_grad_(False).to(conv.weight.device)
# Prepare filters
w_conv = conv.weight.clone().view(conv.out_channels, -1)
w_bn = torch.diag(bn.weight.div(torch.sqrt(bn.eps + bn.running_var)))
fusedconv.weight.copy_(torch.mm(w_bn, w_conv).view(fusedconv.weight.shape))
# Prepare spatial bias
b_conv = torch.zeros(conv.weight.size(0), device=conv.weight.device) if conv.bias is None else conv.bias
b_bn = bn.bias - bn.weight.mul(bn.running_mean).div(torch.sqrt(bn.running_var + bn.eps))
fusedconv.bias.copy_(torch.mm(w_bn, b_conv.reshape(-1, 1)).reshape(-1) + b_bn)
return fusedconv
def model_info(model, verbose=False, imgsz=640):
# Model information. img_size may be int or list, i.e. img_size=640 or img_size=[640, 320]
n_p = sum(x.numel() for x in model.parameters()) # number parameters
n_g = sum(x.numel() for x in model.parameters() if x.requires_grad) # number gradients
if verbose:
print(f"{'layer':>5} {'name':>40} {'gradient':>9} {'parameters':>12} {'shape':>20} {'mu':>10} {'sigma':>10}")
for i, (name, p) in enumerate(model.named_parameters()):
name = name.replace('module_list.', '')
print('%5g %40s %9s %12g %20s %10.3g %10.3g' %
(i, name, p.requires_grad, p.numel(), list(p.shape), p.mean(), p.std()))
try: # FLOPs
p = next(model.parameters())
stride = max(int(model.stride.max()), 32) if hasattr(model, 'stride') else 32 # max stride
im = torch.empty((1, p.shape[1], stride, stride), device=p.device) # input image in BCHW format
flops = thop.profile(deepcopy(model), inputs=(im,), verbose=False)[0] / 1E9 * 2 # stride GFLOPs
imgsz = imgsz if isinstance(imgsz, list) else [imgsz, imgsz] # expand if int/float
fs = f', {flops * imgsz[0] / stride * imgsz[1] / stride:.1f} GFLOPs' # 640x640 GFLOPs
except Exception:
fs = ''
name = Path(model.yaml_file).stem.replace('yolov5', 'YOLOv5') if hasattr(model, 'yaml_file') else 'Model'
LOGGER.info(f"{name} summary: {len(list(model.modules()))} layers, {n_p} parameters, {n_g} gradients{fs}")
def scale_img(img, ratio=1.0, same_shape=False, gs=32): # img(16,3,256,416)
# Scales img(bs,3,y,x) by ratio constrained to gs-multiple
if ratio == 1.0:
return img
h, w = img.shape[2:]
s = (int(h * ratio), int(w * ratio)) # new size
img = F.interpolate(img, size=s, mode='bilinear', align_corners=False) # resize
if not same_shape: # pad/crop img
h, w = (math.ceil(x * ratio / gs) * gs for x in (h, w))
return F.pad(img, [0, w - s[1], 0, h - s[0]], value=0.447) # value = imagenet mean
def copy_attr(a, b, include=(), exclude=()):
# Copy attributes from b to a, options to only include [...] and to exclude [...]
for k, v in b.__dict__.items():
if (len(include) and k not in include) or k.startswith('_') or k in exclude:
continue
else:
setattr(a, k, v)
def smart_optimizer(model, name='Adam', lr=0.001, momentum=0.9, decay=1e-5):
# YOLOv5 3-param group optimizer: 0) weights with decay, 1) weights no decay, 2) biases no decay
g = [], [], [] # optimizer parameter groups
bn = tuple(v for k, v in nn.__dict__.items() if 'Norm' in k) # normalization layers, i.e. BatchNorm2d()
for v in model.modules():
if hasattr(v, 'bias') and isinstance(v.bias, nn.Parameter): # bias (no decay)
g[2].append(v.bias)
if isinstance(v, bn): # weight (no decay)
g[1].append(v.weight)
elif hasattr(v, 'weight') and isinstance(v.weight, nn.Parameter): # weight (with decay)
g[0].append(v.weight)
if name == 'Adam':
optimizer = torch.optim.Adam(g[2], lr=lr, betas=(momentum, 0.999)) # adjust beta1 to momentum
elif name == 'AdamW':
optimizer = torch.optim.AdamW(g[2], lr=lr, betas=(momentum, 0.999), weight_decay=0.0)
elif name == 'RMSProp':
optimizer = torch.optim.RMSprop(g[2], lr=lr, momentum=momentum)
elif name == 'SGD':
optimizer = torch.optim.SGD(g[2], lr=lr, momentum=momentum, nesterov=True)
else:
raise NotImplementedError(f'Optimizer {name} not implemented.')
optimizer.add_param_group({'params': g[0], 'weight_decay': decay}) # add g0 with weight_decay
optimizer.add_param_group({'params': g[1], 'weight_decay': 0.0}) # add g1 (BatchNorm2d weights)
LOGGER.info(f"{colorstr('optimizer:')} {type(optimizer).__name__}(lr={lr}) with parameter groups "
f"{len(g[1])} weight(decay=0.0), {len(g[0])} weight(decay={decay}), {len(g[2])} bias")
return optimizer
def smart_hub_load(repo='ultralytics/yolov5', model='yolov5s', **kwargs):
# YOLOv5 torch.hub.load() wrapper with smart error/issue handling
if check_version(torch.__version__, '1.9.1'):
kwargs['skip_validation'] = True # validation causes GitHub API rate limit errors
if check_version(torch.__version__, '1.12.0'):
kwargs['trust_repo'] = True # argument required starting in torch 0.12
try:
return torch.hub.load(repo, model, **kwargs)
except Exception:
return torch.hub.load(repo, model, force_reload=True, **kwargs)
def smart_resume(ckpt, optimizer, ema=None, weights='yolov5s.pt', epochs=300, resume=True):
# Resume training from a partially trained checkpoint
best_fitness = 0.0
start_epoch = ckpt['epoch'] + 1
if ckpt['optimizer'] is not None:
optimizer.load_state_dict(ckpt['optimizer']) # optimizer
best_fitness = ckpt['best_fitness']
if ema and ckpt.get('ema'):
ema.ema.load_state_dict(ckpt['ema'].float().state_dict()) # EMA
ema.updates = ckpt['updates']
if resume:
assert start_epoch > 0, f'{weights} training to {epochs} epochs is finished, nothing to resume.\n' \
f"Start a new training without --resume, i.e. 'python train.py --weights {weights}'"
LOGGER.info(f'Resuming training from {weights} from epoch {start_epoch} to {epochs} total epochs')
if epochs < start_epoch:
LOGGER.info(f"{weights} has been trained for {ckpt['epoch']} epochs. Fine-tuning for {epochs} more epochs.")
epochs += ckpt['epoch'] # finetune additional epochs
return best_fitness, start_epoch, epochs
class EarlyStopping:
# YOLOv5 simple early stopper
def __init__(self, patience=30):
self.best_fitness = 0.0 # i.e. mAP
self.best_epoch = 0
self.patience = patience or float('inf') # epochs to wait after fitness stops improving to stop
self.possible_stop = False # possible stop may occur next epoch
def __call__(self, epoch, fitness):
if fitness >= self.best_fitness: # >= 0 to allow for early zero-fitness stage of training
self.best_epoch = epoch
self.best_fitness = fitness
delta = epoch - self.best_epoch # epochs without improvement
self.possible_stop = delta >= (self.patience - 1) # possible stop may occur next epoch
stop = delta >= self.patience # stop training if patience exceeded
if stop:
LOGGER.info(f'Stopping training early as no improvement observed in last {self.patience} epochs. '
f'Best results observed at epoch {self.best_epoch}, best model saved as best.pt.\n'
f'To update EarlyStopping(patience={self.patience}) pass a new patience value, '
f'i.e. `python train.py --patience 300` or use `--patience 0` to disable EarlyStopping.')
return stop
class ModelEMA:
""" Updated Exponential Moving Average (EMA) from https://github.com/rwightman/pytorch-image-models
Keeps a moving average of everything in the model state_dict (parameters and buffers)
For EMA details see https://www.tensorflow.org/api_docs/python/tf/train/ExponentialMovingAverage
"""
def __init__(self, model, decay=0.9999, tau=2000, updates=0):
# Create EMA
self.ema = deepcopy(de_parallel(model)).eval() # FP32 EMA
self.updates = updates # number of EMA updates
self.decay = lambda x: decay * (1 - math.exp(-x / tau)) # decay exponential ramp (to help early epochs)
for p in self.ema.parameters():
p.requires_grad_(False)
def update(self, model):
# Update EMA parameters
self.updates += 1
d = self.decay(self.updates)
msd = de_parallel(model).state_dict() # model state_dict
for k, v in self.ema.state_dict().items():
if v.dtype.is_floating_point: # true for FP16 and FP32
v *= d
v += (1 - d) * msd[k].detach()
# assert v.dtype == msd[k].dtype == torch.float32, f'{k}: EMA {v.dtype} and model {msd[k].dtype} must be FP32'
def update_attr(self, model, include=(), exclude=('process_group', 'reducer')):
# Update EMA attributes
copy_attr(self.ema, model, include, exclude)