YOLOv5 v5.0 Release #2762

glenn-jocher · 2021-04-11T17:21:46Z

This release implements YOLOv5-P6 models and retrained YOLOv5-P5 models:

YOLOv5-P5 models (same architecture as v4.0 release): 3 output layers P3, P4, P5 at strides 8, 16, 32, trained at --img 640
YOLOv5-P6 models: 4 output layers P3, P4, P5, P6 at strides 8, 16, 32, 64 trained at --img 1280

Example usage:

# Command Line
python detect.py --weights yolov5m.pt --img 640  # P5 model at 640
python detect.py --weights yolov5m6.pt --img 640  # P6 model at 640
python detect.py --weights yolov5m6.pt --img 1280  # P6 model at 1280

# PyTorch Hub
model = torch.hub.load('ultralytics/yolov5', 'yolov5m6')  #  P6 model
results = model(imgs, size=1280)  # inference at 1280

All model sizes YOLOv5s/m/l/x are now available in P5 and P6 architectures:

python detect.py --weights yolov5s.pt  # P5 models
                           yolov5m.pt
                           yolov5l.pt
                           yolov5x.pt
                           yolov5s6.pt  # P6 models
                           yolov5m6.pt
                           yolov5l6.pt
                           yolov5x6.pt

Notable Updates

YouTube Inference: Direct inference from YouTube videos, i.e. python detect.py --source 'https://youtu.be/NUsoVlDFqZg'. Live streaming videos and normal videos supported. (YouTube Livestream Detection #2752)
AWS Integration: Amazon AWS integration and new AWS Quickstart Guide for simple EC2 instance YOLOv5 training and resuming of interrupted Spot instances. (Amazon AWS EC2 startup and re-startup scripts #2185)
Supervise.ly Integration: New integration with the Supervisely Ecosystem for training and deploying YOLOv5 models with Supervise.ly (Supervisely with YOLOv5 🌟 #2518)
Improved W&B Integration: Allows saving datasets and models directly to Weights & Biases. This allows for --resume directly from W&B (useful for temporary environments like Colab), as well as enhanced visualization tools. See this blog by @AyushExel for details. (Improved W&B integration #2125)

Updated Results

P6 models include an extra P6/64 output layer for detection of larger objects, and benefit the most from training at higher resolution. For this reason we trained all P5 models at 640, and all P6 models at 1280.

YOLOv5-P5 640 Figure (click to expand)

Figure Notes (click to expand)

GPU Speed measures end-to-end time per image averaged over 5000 COCO val2017 images using a V100 GPU with batch size 32, and includes image preprocessing, PyTorch FP16 inference, postprocessing and NMS.
EfficientDet data from google/automl at batch size 8.

April 11, 2021: v5.0 release: YOLOv5-P6 1280 models, AWS, Supervise.ly and YouTube integrations.
January 5, 2021: v4.0 release: nn.SiLU() activations, Weights & Biases logging, PyTorch Hub integration.
August 13, 2020: v3.0 release: nn.Hardswish() activations, data autodownload, native AMP.
July 23, 2020: v2.0 release: improved model definition, training and mAP.

Pretrained Checkpoints

Model	size ^(pixels)	mAP^val 0.5:0.95	mAP^{test 0.5:0.95}	mAP^val 0.5	Speed ^{V100 (ms)}	params ^(M)	FLOPS ^{640 (B)}
YOLOv5s	640	36.7	36.7	55.4	2.0	7.3	17.0
YOLOv5m	640	44.5	44.5	63.3	2.7	21.4	51.3
YOLOv5l	640	48.2	48.2	66.9	3.8	47.0	115.4
YOLOv5x	640	50.4	50.4	68.8	6.1	87.7	218.8

YOLOv5s6	1280	43.3	43.3	61.9	4.3	12.7	17.4
YOLOv5m6	1280	50.5	50.5	68.7	8.4	35.9	52.4
YOLOv5l6	1280	53.4	53.4	71.1	12.3	77.2	117.7
YOLOv5x6	1280	54.4	54.4	72.0	22.4	141.8	222.9

YOLOv5x6 TTA	1280	55.0	55.0	72.0	70.8	-	-

Table Notes (click to expand)

AP^test denotes COCO test-dev2017 server results, all other AP results denote val2017 accuracy.
AP values are for single-model single-scale unless otherwise noted. Reproduce mAP by python test.py --data coco.yaml --img 640 --conf 0.001 --iou 0.65
Speed_GPU averaged over 5000 COCO val2017 images using a GCP n1-standard-16 V100 instance, and includes FP16 inference, postprocessing and NMS. Reproduce speed by python test.py --data coco.yaml --img 640 --conf 0.25 --iou 0.45
All checkpoints are trained to 300 epochs with default settings and hyperparameters (no autoaugmentation).
Test Time Augmentation (TTA) includes reflection and scale augmentation. Reproduce TTA by python test.py --data coco.yaml --img 1536 --iou 0.7 --augment

Changelog

Changes between previous release and this release: v4.0...v5.0
Changes since this release: v5.0...HEAD

Click a section below to expand details:

Implemented Enhancements (26)

Return predictions as json #2703
Single channel image training？ #2609
Images in MPO Format are considered corrupted #2446
Improve Validation Visualization #2384
Add ASFF (three fuse feature layers) int the Head for V5(s,m,l,x) #2348
Dear author, can you provide a visualization scheme for YOLOV5 feature graphs during detect.py? Thank you! #2259
Dataloader #2201
Update Train Custom Data wiki page #2187
Multi-class NMS #2162
💡Idea: Mosaic cropping using segmentation labels #2151
Improving Confusion Matrix Interpretability: FP and FN vectors should be switched to align with Predicted and True axis #2071
Interpreting model YoloV5 by Grad-cam #2065
Output optimal confidence threshold based on PR curve #2048
is it valuable that add --cache-images option to detect.py? #2004
I want to change the anchor box to anchor circles, where do you think the change to be made ? #1987
Support for imgaug #1954
Any plan for Knowledge Distillation? #1762
Is there a wasy to run detections on a video/webcam/rtrsp, etc EVERY x SECONDS? #1742
Can yolov5 support rotated target detection? #1728
Deploying yolov5 to TorchServe (GPU compatible) #1681
Why diffrent colors of bboxs? #1638
Yet another export yolov5 models to ONNX and inference with TensorRT #1597
Rerange the blocks of Focus Layer into row major to be compatible with tensorflow SpaceToDepth #413
YouTube Livestream Detection #2752 (ben-milanko)
Add TransformerLayer, TransformerBlock, C3TR modules #2333 (dingyiwei)
Improved W&B integration #2125 (AyushExel)

Fixed Bugs (73)

it seems that check_wandb_resume don't support multiple input files of images. #2716
ip camera or web camera. error: (-215:Assertion failed) !ss ize.empty() in function 'cv::resize' #2709
Model predict with forward will fail if PIL image does not have filename attribute #2702
❔Question Whenever i try to run my model i run into this error AttributeError: 'NoneType' object has no attribute 'startswith' from wandbutils.py line 161 I wonder why ? Any workaround or fix #2697
coremltools no longer included in docker container #2686
'LoadImages' path handling appears to be broken #2618
CUDA memory leak #2586
UnboundLocalError: local variable 'wandb_logger' referenced before assignment #2562
RuntimeError: CUDA error: CUBLAS_STATUS_INTERNAL_ERROR when calling cublasCreate\(handle\) #2417
CUDNN Mapping Error #2415
Can't train in DDP mode after recent update #2405
a bug about function bbox_iou() #2376
Training got stuck when I used DistributedDataParallel mode but dataParallel mode is useful #2375
Something wrong with fixing ema #2343
Conversion to CoreML fails when running with --batch 2 #2322
The "fitness" function in train.py. #2303
Error "Directory already existed" happen when training with multiple GPUs #2275
self.balance = {3: [4.0, 1.0, 0.4], 4: [4.0, 1.0, 0.25, 0.06], 5: [4.0, 1.0, 0.25, 0.06, .02]}[det.nl] #2255
Cannot run model with URL as argument #2246
Yolov5 crashes with RTSP stream analysis #2226
interruption during evolve #2218
I am a student of Tsinghua University, doing research in Tencent. When I train with yolov5, the following problems appear，Sincerely hope to get help, #2203
Frame Loss in video stream #2196
wandb.ai not logging epochs vs metrics/losses instead uses step #2175
Evolve is leaking files #2142
Issue in torchscript model inference #2129
RuntimeError: CUDA error: device-side assert triggered #2124
In 'evolve' mode, If the original hyp is 0, It will never update #2122
Caching image path #2121
can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first #2106
Error in creating model with Ghost modules #2081
TypeError: int() can't convert non-string with explicit base #2066
[Pytorch Hub] Hub CI is broken with latest master of yolo5 example. #2050
Problems when downloading requirements #2047
detect.py - images always saved #2029
thop and pycocotools shouldn't be hard requirements to train a model #2014
CoreML export failure #2007
loss function like has a bug #1988
CoreML export failure: unexpected number of inputs for node x.2 (_convolution): 13 #1945
torch.nn.modules.module.ModuleAttributeError: 'Hardswish' object has no attribute 'inplace' #1939
runs not logging separately in wandb.ai #1937
wrong batch size after --resume on multiple GPUs #1936
TypeError: int() can't convert non-string with explicit base #1927
RuntimeError: DataLoader worker #1908
Unable to export weights into onnx #1900
CUDA Initialization Warning on Docker when not passing in gpu #1891
Issue with github api rate limiting #1890
wandb: ERROR Error while calling W&B API: Error 1062: Duplicate entry '189160-gbp6y2en' for key 'PRIMARY' (<Response [409]>) #1878
Broken pipe #1859
detection.py #1858
Getting error on loading custom trained model #1856
W&B id is always the same and continue with the old logging. #1851
pytorch1.7 is not completely support.'inplace'! 'inplace'! 'inplace'! #1832
Validation errors are NaN #1804
Error Loading custom model weights with pytorch.hub.load #1788
'cap' object is not self. initialized #1781
ValueError: API key must be 40 characters long, yours was 1 #1777
scipy #1766
error of missing key 'anchors' in hyp.scratch.yaml #1744
mss grab color conversion problem using TorchHub #1735
Video rotation when running detection. #1725
RuntimeError: CUDA out of memory. Tried to allocate 294.00 MiB (GPU 0; 6.00 GiB total capacity; 118.62 MiB already allocated; 4.20 GiB free; 362.00 MiB reserved in total by PyTorch) #1698
Errors on MAC #1690
RuntimeError: DataLoader worker (pid(s) 296430) exited unexpectedly #1675
Non-positive Stride #1671
gbk error. How can I solve it? #1669
CoreML export failure: unexpected number of inputs for node x.2 (_convolution): 13 #1667
RuntimeError: Given groups=1, weight of size [32, 128, 1, 1], expected input[1, 64, 32, 32] to have 128 channels, but got 64 channels instead #1627
segmentation fault #1620
Getting different output sizes when using exported torchscript #1562
some bugs when training #1547
Evolve getting error #1319
AssertionError: Image Not Found ../dataset/images/train/4501.jpeg #195

Closed Issues (42)

Can feed tensor the model #2722
hello, everyone, In order to modify the network more conveniently based on this rep., I restructure the network part, which is divided into backbone, neck, head #2710
Differentiate between normal banner and LED banner #2647
👋 Hello @Wilson-inclaims, thank you for your interest in 🚀 YOLOv5! Please visit our ⭐️ [Tutorials](https://github.com/ultralytics/yolov5/wiki\#tutorials\) to get started, where you can find quickstart guides for simple tasks like [Custom Data Training](https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data\) all the way to advanced concepts like [Hyperparameter Evolution](https://github.com/ultralytics/yolov5/issues/607\). #2516
I got a runtimerror when I run classifier.py to train my own dataset. #2438
RuntimeError: a view of a leaf Variable that requires grad is being used in an in-place operation. #2403
🌟💡 YOLOv5 Study: batch size #2377
export.py export onnx for gpu failed #2365
in _ddp_init_helper expect_sparse_gradient) RuntimeError: Model replicas must have an equal number of parameters. #2311
Custom dataset training using YOLOv5 #2296
label format #2293
MAP NOT PRINTING #2283
Why didn't I get results in my video test? #2277
Label Missing: for images and labels... 203 found, 50 missing, 0 empty, 0 corrupted: 100% #2268
Pytorch Hub inference returns different results than detect.py #2224
yolov5x train.py error #2181
degrees is radians? #2160
AssertionError: Image Not Found #2130
How to load custom trained model to detect sample image? #2097
YOLOv5 installed failed on Macbook M1 #2075
How to set the number of seconds to detect once #2072
Where the changes should be made to detect horizontal line and vertical lines? Can anyone discus elaborately? #2070
Video inference stops after a certain number of frames #2064
Can't YOLOV5 be detected with multithreading? #1979
I want to make a images file what divided images in test.py #1931
different image size w/ torchscript windows c++ #1920
run detect the result ，the Image don't have box #1910
resume problem #1884
Detect source as .txt error #1877
yolov5 v4.0 tensorrt deployment #1874
Hyperparameter Evolution: load dataset every time #1864
Caching images problem #1862
About Release v4.0 #1841
@Glenn best practices for running trained YOLOv5 models in new python environments is to use PyTorch Hub. See PyTorch Hub Tutorial: #1789
yolov5x.pt is not compatible with ./models/yolov5x.yam #1721
Parameter '--device' doesn't work! #1706
Thank you for your issue! #1687
Convert the label format #1652
Autorun not working #1599
When model.model[-1]. export = False in export.py, coreml export failing. Please check. #1491
Error 'AttributeError: 'str' object has no attribute 'get'' at running train.py #1479
Docker image is not working, torch.nn.modules.module.ModuleAttributeError: 'Hardswish' object has no attribute 'inplace' #1327

Merged Pull Requests (172)

Tensorboard model visualization bug fix #2758 (glenn-jocher)
utils/wandb_logging PEP8 reformat #2755 (glenn-jocher)
torch.cuda.amp bug fix #2750 (glenn-jocher)
autocast enable=torch.cuda.is_available() #2748 (glenn-jocher)
Add Hub results.pandas() method #2725 (glenn-jocher)
Update README with collapsable notes #2721 (glenn-jocher)
Fix it seems that check_wandb_resume don't support multiple input files of images. #2716: Add support for list-of-directory data format for wandb #2719 (AyushExel)
Updated filename attributes for YOLOv5 Hub BytesIO #2718 (glenn-jocher)
Updated filename attributes for YOLOv5 Hub results #2708 (glenn-jocher)
pip install coremltools onnx #2690 (glenn-jocher)
autoShape forward im = np.asarray(im) # to numpy #2689 (glenn-jocher)
PyTorch Hub model.save() increment as runs/hub/exp #2684 (glenn-jocher)
Fix: wandbutils.py line 161 AttributeError: 'NoneType' object has no attribute 'startswith' #2674 #2683 (AyushExel)
Update README with Tips for Best Results tutorial #2682 (glenn-jocher)
Disallow resume for dataset uploading #2657 (AyushExel)
Speed profiling improvements #2648 (glenn-jocher)
Add tqdm pbar.close() #2644 (zzttqu)
PyTorch Hub amp.autocast() inference #2641 (glenn-jocher)
PyTorch Hub custom model to CUDA device fix #2636 (glenn-jocher)
Improve git_describe() fix 1 #2635 (glenn-jocher)
Fix: evolve with wandb #2634 (AyushExel)
Improve git_describe() #2633 (glenn-jocher)
FROM nvcr.io/nvidia/pytorch:21.03-py3 #2623 (glenn-jocher)
Remove conflicting nvidia-tensorboard package #2622 (glenn-jocher)
Update Detections() self.n comment #2620 (glenn-jocher)
Update detections() self.t = tuple() #2617 (glenn-jocher)
Create date_modified() #2616 (glenn-jocher)
Added '.mpo' to supported image formats #2615 (maxupp)
Fix Indentation in test.py #2614 (AyushExel)
Update git_describe() for remote dir usage #2606 (glenn-jocher)
Remove Cython from requirements.txt #2604 (glenn-jocher)
resume.py fix DDP typo #2603 (glenn-jocher)
Update segment2box() comment #2600 (glenn-jocher)
Save webcam results, add --nosave option #2598 (glenn-jocher)
YOLOv5 PyTorch Hub models >> check_requirements() #2592 (glenn-jocher)
YOLOv5 PyTorch Hub models >> check_requirements() #2591 (glenn-jocher)
YOLOv5 PyTorch Hub models >> check_requirements() #2588 (glenn-jocher)
W&B DDP fix 2 #2587 (glenn-jocher)
W&B resume ddp from run link fix #2579 (AyushExel)
YOLOv5 PyTorch Hub models >> check_requirements() #2577 (glenn-jocher)
Update tensorboard>=2.4.1 #2576 (glenn-jocher)
Enhanced check_requirements() with auto-install #2575 (glenn-jocher)
W&B DDP fix #2574 (AyushExel)
check_requirements() exclude pycocotools, thop #2571 (glenn-jocher)
Update Detections() times=None #2570 (glenn-jocher)
Add opencv-contrib-python to requirements.txt #2564 (youngjinshin)
Supervisely Ecosystem #2519 (mkolomeychenko)
add option to disable half precision when testing #2507 (bfineran)
PyTorch Hub models default to CUDA:0 if available #2472 (glenn-jocher)
Scipy kmeans-robust autoanchor update #2470 (glenn-jocher)
Be able to create dataset from annotated images only #2466 (kinoute)
autoShape() speed profiling update #2460 (glenn-jocher)
Add autoShape() speed profiling #2459 (glenn-jocher)
CVPR 2021 Argoverse-HD autodownload curl #2455 (glenn-jocher)
labels.jpg class names #2454 (glenn-jocher)
Update test.py --task train val study #2453 (glenn-jocher)
Integer printout #2450 (glenn-jocher)
DDP after autoanchor reorder #2421 (glenn-jocher)
CVPR 2021 Argoverse-HD autodownload fix #2418 (glenn-jocher)
CVPR 2021 Argoverse-HD dataset autodownload support #2400 (karthiksharma98)
GCP sudo docker userdata.sh #2393 (glenn-jocher)
AWS wait && echo "All tasks done." #2391 (glenn-jocher)
bbox_iou() stability and speed improvements #2385 (glenn-jocher)
image weights compatible faster random index generator v2 for mosaic … #2383 (developer0hye)
ENV HOME=/usr/src/app #2382 (glenn-jocher)
--no-cache notebook #2381 (glenn-jocher)
Resume with custom anchors fix #2361 (glenn-jocher)
Anchor override for training from scratch #2350 (glenn-jocher)
faster random index generator for mosaic augmentation #2345 (developer0hye)
Add --label-smoothing eps argument to train.py (default 0.0) #2344 (ptran1203)
FROM nvcr.io/nvidia/pytorch:21.02-py3 #2341 (glenn-jocher)
EMA bug fix 2 #2330 (glenn-jocher)
remove TTA 1 pixel offset #2325 (glenn-jocher)
Update Dockerfile install htop #2320 (glenn-jocher)
Update test.py #2319 (glenn-jocher)
final_epoch EMA bug fix #2317 (glenn-jocher)
Fix labels being missed when image extension appears twice in filename #2300 (idenc)
W&B entity support #2298 (toretak)
GPU export options #2297 (toretak)
Improved model+EMA checkpointing fix #2295 (glenn-jocher)
Improved model+EMA checkpointing #2292 (glenn-jocher)
Update train.py #2290 (glenn-jocher)
Amazon AWS EC2 startup and re-startup scripts patch #2282 (glenn-jocher)
FLOPS min stride 32 #2276 (xiaowo1996)
Update Dockerfile with apt install zip #2274 (glenn-jocher)
Update greetings.yml for auto-rebase on PR #2272 (glenn-jocher)
Update minimum stride to 32 #2266 (glenn-jocher)
Robust objectness loss balancing #2256 (glenn-jocher)
Update inference default to multi_label=False #2252 (glenn-jocher)
Improved hubconf.py CI tests #2251 (glenn-jocher)
YOLOv5 Hub URL inference bug fix #2250 (glenn-jocher)
Unified hub and detect.py box and labels plotting #2243 (kinoute)
Add isdocker() #2232 (glenn-jocher)
Add check_imshow() #2231 (glenn-jocher)
Update CI badge #2230 (glenn-jocher)
Update yolo.py channel array #2223 (glenn-jocher)
LoadStreams() frame loss bug fix #2222 (glenn-jocher)
TTA augument boxes one pixel shifted in de-flip ud and lr #2219 (VdLMV)
Dynamic ONNX engine generation #2208 (aditya-dl)
YOLOv5 PyTorch Hub results.save() method retains filenames #2194 (dan0nchik)
YOLOv5 Segmentation Dataloader Updates #2188 (glenn-jocher)
Amazon AWS EC2 startup and re-startup scripts #2185 (glenn-jocher)
PyTorch Hub results.save('path/to/dir') #2179 (glenn-jocher)
Changed socket port and added timeout #2176 (NanoCode012)
Update utils/datasets.py to support .webp files #2174 (Transigent)
Update requirements.txt #2173 (glenn-jocher)
Update detect.py #2167 (ab-101)
Update data-autodownload background tasks #2154 (glenn-jocher)
Linear LR scheduler option #2150 (glenn-jocher)
Update train.py batch_size * 2 #2149 (glenn-jocher)
Update train.py test batch_size #2148 (glenn-jocher)
LoadImages() pathlib update #2140 (glenn-jocher)
Unique *.cache filenames fix #2134 (train255)
Making inference thread-safe by avoiding mutable state in Detect class #2120 (olehb)
Make normalized confusion matrix more interpretable #2114 (rbavery)
Update plot_study() #2112 (glenn-jocher)
Update test.py --task speed and study #2099 (glenn-jocher)
Add variable-stride inference support #2091 (glenn-jocher)
Add Kaggle badge #2090 (glenn-jocher)
Add Amazon Deep Learning AMI environment #2085 (glenn-jocher)
Add YOLOv5-P6 models #2083 (glenn-jocher)
GhostConv update #2082 (glenn-jocher)
W&B epoch logging update #2073 (glenn-jocher)
Update to colors.TABLEAU_COLORS #2069 (glenn-jocher)
Update run-once lines #2058 (glenn-jocher)
Metric-Confidence plots feature addition #2057 (glenn-jocher)
Add histogram equalization fcn #2049 (glenn-jocher)
Confusion matrix native image-space fix #2046 (ramonhollands)
Check im.format during dataset caching #2042 (glenn-jocher)
Add exclude tuple to check_requirements() #2041 (glenn-jocher)
data-autodownload background tasks #2034 (glenn-jocher)
Docker pyYAML>=5.3.1 fix #2031 (glenn-jocher)
PyYAML==5.4.1 #2030 (glenn-jocher)
Update autoshape .print() and .save() #2022 (glenn-jocher)
Update requirements.txt #2021 (glenn-jocher)
Update general.py check_git_status() fix #2020 (glenn-jocher)
Update inference multiple-counting #2019 (glenn-jocher)
Update ci-testing.yml --img 128 #2018 (glenn-jocher)
Update google_utils.py attempt_download() fix #2017 (glenn-jocher)
Update Dockerfile #2016 (glenn-jocher)
check_git_status() Windows fix #2015 (glenn-jocher)
--verbose on final_epoch #1997 (glenn-jocher)
Add xywhn2xyxy() #1983 (glenn-jocher)
Update Dockerfile #1982 (glenn-jocher)
check_git_status() asserts #1977 (glenn-jocher)
Update train.py #1972 (Anon-Artist)
Update autoanchor.py #1971 (Anon-Artist)
Update yolo.py #1970 (Anon-Artist)
Update test.py #1969 (Anon-Artist)
Update plots.py #1968 (Anon-Artist)
check_git_status() when not exist /workspace #1966 (glenn-jocher)
Security Fix for Arbitrary Code Execution - huntr.dev #1962 (huntr-helper)
prevent check_git_status() in docker images #1951 (glenn-jocher)
Add ComputeLoss() class #1950 (glenn-jocher)
W&B mosaic log bug fix #1949 (glenn-jocher)
Start setting up the improved W&B integration #1948 (AyushExel)
W&B log epoch #1946 (glenn-jocher)
Daemon thread mosaic plots fix #1943 (glenn-jocher)
Fix batch-size on resume for multi-gpu #1942 (NanoCode012)
Add nn.SiLU inplace in attempt_load() #1940 (1991wangliang)
GitHub API rate limit fallback #1930 (glenn-jocher)
check_git_status() bug fix #1925 (glenn-jocher)
check_git_status() improvements #1916 (glenn-jocher)
Colorstr() updates #1909 (glenn-jocher)
PyTorch Hub results.render() #1897 (glenn-jocher)
Docker CUDA warning fix #1895 (glenn-jocher)
GitHub API rate limit fix #1894 (glenn-jocher)
Add colorstr() #1887 (glenn-jocher)
auto-verbose if nc <= 20 #1869 (glenn-jocher)
actions/stale@v3 #1868 (glenn-jocher)
Add check_requirements() #1853 (glenn-jocher)
W&B ID reset on training completion #1852 (TommyZihao)

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhancements and clarifications in YOLOv5 documentation and models.

📊 Key Changes

Updated the main image in the README.md to a new one, and added an expandable section featuring the YOLOv5-P5 640 model figure.
Provided a command for reproducing study results directly in the documentation.
Noted the v5.0 release, which included new YOLOv5-P6 1280 model variants and integrations with AWS, Supervise.ly, and YouTube.
Adjusted pretrained checkpoints table to streamline model details and added new performance stats for the models, including a new YOLOv5s6 variant.
Simplified and updated the detect.py description to clarify that it downloads models from the latest release and saves results to a specific path.
Refactored the hubconf.py by moving the custom function and updating the create functions for each model variant to be more succinct and focus on creation and configuration.
Updated the plot_study_txt function in utils/plots.py to fix the plotting range and updated the reference to the YOLOv5s6 model family.

🎯 Purpose & Impact

Documentation: Improves understanding of the repository's capabilities and recent changes, making it easier for both users and contributors to engage with the project.
Model Performance: The new model stats and variants expand the options for users, allowing better performance tuning for different computational environments.
Code Clarity: The cleanup in hubconf.py makes it easier for developers to understand the structure of model loading and instantiation.
Visualization Updates: Adjustments to the plotting functions aid in better visual representation of model performances.
Overall Impact: The PR will potentially increase user adoption and improve the user experience by providing clear, updated information and better-performing models.

AyushExel · 2021-04-11T17:26:00Z

Awesome 😎

Edwardmark · 2021-04-16T09:15:55Z

@glenn-jocher What does the B in flops mean?

glenn-jocher · 2021-04-16T19:08:03Z

@Edwardmark M for Million (1E6) and B for Billion (1E9), i.e. 1 Billion FLOPS or 1 GFLOP

@Pro100rus32

* Update yolo.py with yaml.SafeLoader (ultralytics#1970) * Update autoanchor.py with yaml.SafeLoader (ultralytics#1971) * Update train.py with yaml.SafeLoader (ultralytics#1972) * check_git_status() asserts (ultralytics#1977) * Update Dockerfile (ultralytics#1982) * Add xywhn2xyxy() (ultralytics#1983) * verbose on final_epoch (ultralytics#1997) * check_git_status() Windows fix (ultralytics#2015) * check_git_status() Windows fix * Update general.py * Update general.py * Update general.py * Update general.py * Update general.py * Update general.py * Update Dockerfile (ultralytics#2016) * Update google_utils.py (ultralytics#2017) * Update ci-testing.yml (ultralytics#2018) * Update inference multiple-counting (ultralytics#2019) * Update inference multiple-counting * update github check * Update general.py check_git_status() fix (ultralytics#2020) * Update autoshape .print() and .save() (ultralytics#2022) * Update requirements.txt (ultralytics#2021) * Update requirements.txt * Update ci-testing.yml * Update hubconf.py * PyYAML==5.4.1 (ultralytics#2030) * Docker pyYAML>=5.3.1 fix (ultralytics#2031) * data-autodownload background tasks (ultralytics#2034) * Check im.format during dataset caching (ultralytics#2042) * Check im.format during dataset caching * Update datasets.py * Confusion matrix native image-space fix (ultralytics#2046) Make sure the labels and predictions are equally scaled on confusion_matrix.process_batch * Add histogram equalization fcn (ultralytics#2049) * W&B log epoch (ultralytics#1946) * W&B log epoch * capitalize * W&B log epoch * capitalize * Update train.py New try using https://docs.wandb.ai/library/log#incremental-logging * Update train.py * Update test.py * Update train.py * Update plots.py * Update train.py * Update train.py * label plot step -1 * update * update * update * update * update * update * Update train.py * Update train.py * Add 'exclude' tuple to check_requirements() (ultralytics#2041) * Update run-once lines (ultralytics#2058) * Metric-Confidence plots feature addition (ultralytics#2057) * Metric-Confidence plots feature addition * cleanup * Metric-Confidence plots feature addition * cleanup * Update run-once lines * cleanup * save all 4 curves to wandb * Update to colors.TABLEAU_COLORS (ultralytics#2069) * W&B epoch logging update (ultralytics#2073) * GhostConv update (ultralytics#2082) * Add YOLOv5-P6 models (ultralytics#2083) * Update tutorial.ipynb * Add Amazon Deep Learning AMI environment (ultralytics#2085) * Update greetings.yml * Update README.md * Add Kaggle badge (ultralytics#2090) * Update README.md * Update greetings.yml * Created using Colaboratory * Add Kaggle badge (ultralytics#2090) * Add variable-stride inference support (ultralytics#2091) * Update test.py --task speed and study (ultralytics#2099) * Add --speed benchmark * test range 256 - 1536 * update * update * update * update * Update plot_study() (ultralytics#2112) * Start setup for improved W&B integration (ultralytics#1948) * Add helper functions for wandb and artifacts * cleanup * Reorganize files * Update wandb_utils.py * Update log_dataset.py We can remove this code, as the giou hyp has been deprecated for a while now. * Reorganize and update dataloader call * yaml.SafeLoader * PEP8 reformat * remove redundant checks * Add helper functions for wandb and artifacts * cleanup * Reorganize files * Update wandb_utils.py * Update log_dataset.py We can remove this code, as the giou hyp has been deprecated for a while now. * Reorganize and update dataloader call * yaml.SafeLoader * PEP8 reformat * remove redundant checks * Update util files * Update wandb_utils.py * Remove word size * Change path of labels.zip * remove unused imports * remove --rect * log_dataset.py cleanup * log_dataset.py cleanup2 * wandb_utils.py cleanup * remove redundant id_count * wandb_utils.py cleanup2 * rename cls * use pathlib for zip * rename dataloader to dataset * Change import order * Remove redundant code * remove unused import * remove unused imports Co-authored-by: Glenn Jocher <[email protected]> * LoadImages() pathlib update (ultralytics#2140) * Unique *.cache filenames fix (ultralytics#2134) * fix ultralytics#2121 * Update test.py * Update train.py * Update autoanchor.py * Update datasets.py * Update log_dataset.py * Update datasets.py Co-authored-by: Glenn Jocher <[email protected]> * Update train.py test batch_size (ultralytics#2148) * Update train.py * Update loss.py * Update train.py (ultralytics#2149) * Linear LR scheduler option (ultralytics#2150) * Linear LR scheduler option * Update train.py * Update data-autodownload background tasks (ultralytics#2154) * Update get_coco.sh * Update get_voc.sh * Update detect.py (ultralytics#2167) Without this cv2.imshow opens a window but nothing is visible * Update requirements.txt (ultralytics#2173) * Update utils/datasets.py to support .webp files (ultralytics#2174) Simply added 'webp' as an image format to the img_formats array so that webp image files can be used as training data. * Changed socket port and added timeout (ultralytics#2176) * PyTorch Hub results.save('path/to/dir') (ultralytics#2179) * YOLOv5 Segmentation Dataloader Updates (ultralytics#2188) * Update C3 module * Update C3 module * Update C3 module * Update C3 module * update * update * update * update * update * update * update * update * update * updates * updates * updates * updates * updates * updates * updates * updates * updates * updates * update * update * update * update * updates * updates * updates * updates * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update datasets * update * update * update * update attempt_downlaod() * merge * merge * update * update * update * update * update * update * update * update * update * update * parameterize eps * comments * gs-multiple * update * max_nms implemented * Create one_cycle() function * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * GitHub API rate limit fix * update * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * astuple * epochs * update * update * ComputeLoss() * update * update * update * update * update * update * update * update * update * update * update * merge * merge * merge * merge * update * update * update * update * commit=tag == tags[-1] * Update cudnn.benchmark * update * update * update * updates * updates * updates * updates * updates * updates * updates * update * update * update * update * update * mosaic9 * update * update * update * update * update * update * institute cache versioning * only display on existing cache * reverse cache exists booleans * Created using Colaboratory * YOLOv5 PyTorch Hub results.save() method retains filenames (ultralytics#2194) * save results with name * debug * save original imgs names * Update common.py Co-authored-by: Glenn Jocher <[email protected]> * TTA augument boxes one pixel shifted in de-flip ud and lr (ultralytics#2219) * TTA augument boxes one pixel shifted in de-flip ud and lr * PEP8 reformat Co-authored-by: Jaap van de Loosdrecht <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> * LoadStreams() frame loss bug fix (ultralytics#2222) * Update yolo.py channel array (ultralytics#2223) * Add check_imshow() (ultralytics#2231) * Add check_imshow() * Update general.py * Update general.py * Update CI badge (ultralytics#2230) * Add isdocker() (ultralytics#2232) * Add isdocker() * Update general.py * Update general.py * YOLOv5 Hub URL inference bug fix (ultralytics#2250) * Update common.py * Update common.py * Update common.py * Improved hubconf.py CI tests (ultralytics#2251) * Unified hub and detect.py box and labels plotting (ultralytics#2243) * reset head * Update inference default to multi_label=False (ultralytics#2252) * Update inference default to multi_label=False * bug fix * Update plots.py * Update plots.py * Robust objectness loss balancing (ultralytics#2256) * Created using Colaboratory * Update minimum stride to 32 (ultralytics#2266) * Dynamic ONNX engine generation (ultralytics#2208) * add: dynamic onnx export * delete: test onnx inference * fix dynamic output axis * Code reduction * fix: dynamic output axes, dynamic input naming * Remove fixed axes Co-authored-by: Shivam Swanrkar <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> * Update greetings.yml for auto-rebase on PR (ultralytics#2272) * Update Dockerfile with apt install zip (ultralytics#2274) * FLOPS min stride 32 (ultralytics#2276) Signed-off-by: xiaowo1996 <[email protected]> * Update README.md * Amazon AWS EC2 startup and re-startup scripts (ultralytics#2185) * Amazon AWS EC2 startup and re-startup scripts * Create resume.py * cleanup * Amazon AWS EC2 startup and re-startup scripts (ultralytics#2282) * Update train.py (ultralytics#2290) * Update train.py * Update train.py * Update train.py * Update train.py * Create train.py * Improved model+EMA checkpointing (ultralytics#2292) * Enhanced model+EMA checkpointing * update * bug fix * bug fix 2 * always save optimizer * ema half * remove model.float() * model half * carry ema/model in fp32 * rm model.float() * both to float always * cleanup * cleanup * Improved model+EMA checkpointing 2 (ultralytics#2295) * Fix labels being missed when image extension appears twice in filename (ultralytics#2300) * W&B entity support (ultralytics#2298) * W&B entity support * shorten wandb_entity to entity Co-authored-by: Jan Hajek <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> * Update yolo.py (ultralytics#2120) * Avoid mutable state in Detect * LoadImages() pathlib update (ultralytics#2140) * Unique *.cache filenames fix (ultralytics#2134) * fix ultralytics#2121 * Update test.py * Update train.py * Update autoanchor.py * Update datasets.py * Update log_dataset.py * Update datasets.py Co-authored-by: Glenn Jocher <[email protected]> * Update train.py test batch_size (ultralytics#2148) * Update train.py * Update loss.py * Update train.py (ultralytics#2149) * Linear LR scheduler option (ultralytics#2150) * Linear LR scheduler option * Update train.py * Update data-autodownload background tasks (ultralytics#2154) * Update get_coco.sh * Update get_voc.sh * Update detect.py (ultralytics#2167) Without this cv2.imshow opens a window but nothing is visible * Update requirements.txt (ultralytics#2173) * Update utils/datasets.py to support .webp files (ultralytics#2174) Simply added 'webp' as an image format to the img_formats array so that webp image files can be used as training data. * Changed socket port and added timeout (ultralytics#2176) * PyTorch Hub results.save('path/to/dir') (ultralytics#2179) * YOLOv5 Segmentation Dataloader Updates (ultralytics#2188) * Update C3 module * Update C3 module * Update C3 module * Update C3 module * update * update * update * update * update * update * update * update * update * updates * updates * updates * updates * updates * updates * updates * updates * updates * updates * update * update * update * update * updates * updates * updates * updates * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update datasets * update * update * update * update attempt_downlaod() * merge * merge * update * update * update * update * update * update * update * update * update * update * parameterize eps * comments * gs-multiple * update * max_nms implemented * Create one_cycle() function * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * update * GitHub API rate limit fix * update * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * ComputeLoss * astuple * epochs * update * update * ComputeLoss() * update * update * update * update * update * update * update * update * update * update * update * merge * merge * merge * merge * update * update * update * update * commit=tag == tags[-1] * Update cudnn.benchmark * update * update * update * updates * updates * updates * updates * updates * updates * updates * update * update * update * update * update * mosaic9 * update * update * update * update * update * update * institute cache versioning * only display on existing cache * reverse cache exists booleans * Created using Colaboratory * YOLOv5 PyTorch Hub results.save() method retains filenames (ultralytics#2194) * save results with name * debug * save original imgs names * Update common.py Co-authored-by: Glenn Jocher <[email protected]> * TTA augument boxes one pixel shifted in de-flip ud and lr (ultralytics#2219) * TTA augument boxes one pixel shifted in de-flip ud and lr * PEP8 reformat Co-authored-by: Jaap van de Loosdrecht <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> * LoadStreams() frame loss bug fix (ultralytics#2222) * Update yolo.py channel array (ultralytics#2223) * Add check_imshow() (ultralytics#2231) * Add check_imshow() * Update general.py * Update general.py * Update CI badge (ultralytics#2230) * Add isdocker() (ultralytics#2232) * Add isdocker() * Update general.py * Update general.py * YOLOv5 Hub URL inference bug fix (ultralytics#2250) * Update common.py * Update common.py * Update common.py * Improved hubconf.py CI tests (ultralytics#2251) * Unified hub and detect.py box and labels plotting (ultralytics#2243) * reset head * Update inference default to multi_label=False (ultralytics#2252) * Update inference default to multi_label=False * bug fix * Update plots.py * Update plots.py * Robust objectness loss balancing (ultralytics#2256) * Created using Colaboratory * Update minimum stride to 32 (ultralytics#2266) * Dynamic ONNX engine generation (ultralytics#2208) * add: dynamic onnx export * delete: test onnx inference * fix dynamic output axis * Code reduction * fix: dynamic output axes, dynamic input naming * Remove fixed axes Co-authored-by: Shivam Swanrkar <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> * Update greetings.yml for auto-rebase on PR (ultralytics#2272) * Update Dockerfile with apt install zip (ultralytics#2274) * FLOPS min stride 32 (ultralytics#2276) Signed-off-by: xiaowo1996 <[email protected]> * Update README.md * Amazon AWS EC2 startup and re-startup scripts (ultralytics#2185) * Amazon AWS EC2 startup and re-startup scripts * Create resume.py * cleanup * Amazon AWS EC2 startup and re-startup scripts (ultralytics#2282) * Update train.py (ultralytics#2290) * Update train.py * Update train.py * Update train.py * Update train.py * Create train.py * Improved model+EMA checkpointing (ultralytics#2292) * Enhanced model+EMA checkpointing * update * bug fix * bug fix 2 * always save optimizer * ema half * remove model.float() * model half * carry ema/model in fp32 * rm model.float() * both to float always * cleanup * cleanup * Improved model+EMA checkpointing 2 (ultralytics#2295) * Fix labels being missed when image extension appears twice in filename (ultralytics#2300) * W&B entity support (ultralytics#2298) * W&B entity support * shorten wandb_entity to entity Co-authored-by: Jan Hajek <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> * Avoid mutable state in Detect * Update yolo and remove .to(device) Co-authored-by: Oleg Boiko <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> Co-authored-by: train255 <[email protected]> Co-authored-by: ab-101 <[email protected]> Co-authored-by: Transigent <[email protected]> Co-authored-by: NanoCode012 <[email protected]> Co-authored-by: Daniel Khromov <[email protected]> Co-authored-by: VdLMV <[email protected]> Co-authored-by: Jaap van de Loosdrecht <[email protected]> Co-authored-by: Yann Defretin <[email protected]> Co-authored-by: Aditya Lohia <[email protected]> Co-authored-by: Shivam Swanrkar <[email protected]> Co-authored-by: xiaowo1996 <[email protected]> Co-authored-by: Iden Craven <[email protected]> Co-authored-by: Jan Hajek <[email protected]> Co-authored-by: Jan Hajek <[email protected]> * final_epoch EMA bug fix (ultralytics#2317) * Update test.py (ultralytics#2319) * Update Dockerfile install htop (ultralytics#2320) * remove TTA 1 pixel offset (ultralytics#2325) * EMA bug fix 2 (ultralytics#2330) * EMA bug fix 2 * update * FROM nvcr.io/nvidia/pytorch:21.02-py3 (ultralytics#2341) * Confusion matrix background axis swap (ultralytics#2114) * Created using Colaboratory * Anchor override (ultralytics#2350) * Resume with custom anchors fix (ultralytics#2361) * Resume with custom anchors fix * Update train.py * Faster random index generator for mosaic augmentation (ultralytics#2345) * faster random index generator for mosaic augementation We don't need to access list to generate random index It makes augmentation slower. * Update datasets.py Co-authored-by: Glenn Jocher <[email protected]> * --no-cache notebook (ultralytics#2381) * ENV HOME=/usr/src/app (ultralytics#2382) Set HOME environment variable per Binder requirements. https://github.com/binder-examples/minimal-dockerfile * image weights compatible faster random index generator v2 for mosaic augmentation (ultralytics#2383) image weights compatible faster random index generator v2 for mosaic augmentation * GPU export options (ultralytics#2297) * option for skip last layer and cuda export support * added parameter device * fix import * cleanup 1 * cleanup 2 * opt-in grid --grid will export with grid computation, default export will skip grid (same as current) * default --device cpu GPU export causes ONNX and CoreML errors. Co-authored-by: Jan Hajek <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> * bbox_iou() stability and speed improvements (ultralytics#2385) * AWS wait && echo "All tasks done." (ultralytics#2391) * GCP sudo docker userdata.sh (ultralytics#2393) * GCP sudo docker * cleanup * CVPR 2021 Argoverse-HD dataset autodownload support (ultralytics#2400) * added argoverse-download ability * bugfix * add support for Argoverse dataset * Refactored code * renamed to argoverse-HD * unzip -q and YOLOv5 small cleanup items * add image counts Co-authored-by: Kartikeya Sharma <[email protected]> Co-authored-by: Kartikeya Sharma <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> * CVPR 2021 Argoverse-HD autodownload fix (ultralytics#2418) * DDP after autoanchor reorder (ultralytics#2421) * Integer printout (ultralytics#2450) * Integer printout * test.py 'Labels' * Update train.py * Update test.py --task train val study (ultralytics#2453) * Update test.py --task train val study * update argparser --task * labels.jpg class names (ultralytics#2454) * labels.png class names * fontsize=10 * CVPR 2021 Argoverse-HD autodownload curl (ultralytics#2455) curl preferred over wget for slightly better cross platform compatibility (i.e. out of the box macos compatible). * Add autoShape() speed profiling (ultralytics#2459) * Add autoShape() speed profiling * Update common.py * Create README.md * Update hubconf.py * cleanuip * autoShape() speed profiling update (ultralytics#2460) * Update tutorial.ipynb * Created using Colaboratory * Update autosplit() with annotated_only option (ultralytics#2466) * Be able to create dataset from annotated images only Add the ability to create a dataset/splits only with images that have an annotation file, i.e a .txt file, associated to it. As we talked about this, the absence of a txt file could mean two things: * either the image wasn't yet labelled by someone, * either there is no object to detect. When it's easy to create small datasets, when you have to create datasets with thousands of images (and more coming), it's hard to track where you at and you don't want to wait to have all of them annotated before starting to train. Which means some images would lack txt files and annotations, resulting in label inconsistency as you say in ultralytics#2313. By adding the annotated_only argument to the function, people could create, if they want to, datasets/splits only with images that were labelled, for sure. * Cleanup and update print() Co-authored-by: Glenn Jocher <[email protected]> * Scipy kmeans-robust autoanchor update (ultralytics#2470) Fix for ultralytics#2394 * PyTorch Hub models default to CUDA:0 if available (ultralytics#2472) * PyTorch Hub models default to CUDA:0 if available * device as string bug fix * Created using Colaboratory * Improved W&B integration (ultralytics#2125) * Init Commit * new wandb integration * Update * Use data_dict in test * Updates * Update: scope of log_img * Update: scope of log_img * Update * Update: Fix logging conditions * Add tqdm bar, support for .txt dataset format * Improve Result table Logger * Init Commit * new wandb integration * Update * Use data_dict in test * Updates * Update: scope of log_img * Update: scope of log_img * Update * Update: Fix logging conditions * Add tqdm bar, support for .txt dataset format * Improve Result table Logger * Add dataset creation in training script * Change scope: self.wandb_run * Add wandb-artifact:// natively you can now use --resume with wandb run links * Add suuport for logging dataset while training * Cleanup * Fix: Merge conflict * Fix: CI tests * Automatically use wandb config * Fix: Resume * Fix: CI * Enhance: Using val_table * More resume enhancement * FIX : CI * Add alias * Get useful opt config data * train.py cleanup * Cleanup train.py * more cleanup * Cleanup| CI fix * Reformat using PEP8 * FIX:CI * rebase * remove uneccesary changes * remove uneccesary changes * remove uneccesary changes * remove unecessary chage from test.py * FIX: resume from local checkpoint * FIX:resume * FIX:resume * Reformat * Performance improvement * Fix local resume * Fix local resume * FIX:CI * Fix: CI * Imporve image logging * (:(:Redo CI tests:):) * Remember epochs when resuming * Remember epochs when resuming * Update DDP location Potential fix for ultralytics#2405 * PEP8 reformat * 0.25 confidence threshold * reset train.py plots syntax to previous * reset epochs completed syntax to previous * reset space to previous * remove brackets * reset comment to previous * Update: is_coco check, remove unused code * Remove redundant print statement * Remove wandb imports * remove dsviz logger from test.py * Remove redundant change from test.py * remove redundant changes from train.py * reformat and improvements * Fix typo * Add tqdm tqdm progress when scanning files, naming improvements Co-authored-by: Glenn Jocher <[email protected]> * Update Detections() times=None (ultralytics#2570) Fix for results.tolist() method breaking after YOLOv5 Hub profiling PRshttps://github.com/ultralytics/pull/2460 ultralytics#2459 and * check_requirements() exclude pycocotools, thop (ultralytics#2571) Exclude non-critical packages from dependency checks in detect.py. pycocotools and thop in particular are not required for inference. Issue first raised in ultralytics#1944 and also raised in ultralytics#2556 * W&B DDP fix (ultralytics#2574) * Enhanced check_requirements() with auto-install (ultralytics#2575) * Update check_requirements() with auto-install This PR builds on an idea I had to automatically install missing dependencies rather than simply report an error message. YOLOv5 should now 1) display all dependency issues and not simply display the first missing dependency, and 2) attempt to install/update each missing/VersionConflict package. * cleanup * cleanup 2 * Check requirements.txt file exists * cleanup 3 * Update tensorboard>=2.4.1 (ultralytics#2576) * Update tensorboard>=2.4.1 Update tensorboard version to attempt to address ultralytics#2573 (tensorboard logging fail in Docker image). * cleanup * YOLOv5 PyTorch Hub models >> check_requirements() (ultralytics#2577) * Update hubconf.py with check_requirements() Dependency checks have been missing from YOLOv5 PyTorch Hub model loading, causing errors in some cases when users are attempting to import hub models in unsupported environments. This should examine the YOLOv5 requirements.txt file and pip install any missing or version-conflict packages encountered. This is highly experimental (!), please let us know if this creates problems in your custom workflows. * Update hubconf.py * W&B DDP fix 2 (ultralytics#2587) Revert unintentional change to test batch sizes caused by PR ultralytics#2125 * YOLOv5 PyTorch Hub models >> check_requirements() (ultralytics#2588) * YOLOv5 PyTorch Hub models >> check_requirements() Update YOLOv5 PyTorch Hub requirements.txt path to cache path. * Update hubconf.py * YOLOv5 PyTorch Hub models >> check_requirements() (ultralytics#2591) Prints 'Please restart runtime or rerun command for update to take effect.' following package auto-install to inform users to restart/rerun. * YOLOv5 PyTorch Hub models >> check_requirements() (ultralytics#2592) Improved user-feedback following requirements auto-update. * Supervisely Ecosystem (ultralytics#2519) guide describes YOLOv5 apps collection in Supervisely Ecosystem * Save webcam results, add --nosave option (ultralytics#2598) This updates the default detect.py behavior to automatically save all inference images/videos/webcams unless the new argument --nosave is used (python detect.py --nosave) or unless a list of streaming sources is passed (python detect.py --source streams.txt) * Update segment2box() comment (ultralytics#2600) * resume.py typo (ultralytics#2603) * Remove Cython from requirements.txt (ultralytics#2604) Cython should be a dependency of the remaining packages in requirements.txt, so should be installed anyway even if not a direct requirement. * Update git_describe() for remote dir usage (ultralytics#2606) * Add '*.mpo' to supported image formats (ultralytics#2615) Co-authored-by: Max Uppenkamp <[email protected]> * Create date_modified() (ultralytics#2616) Updated device selection string with fallback for non-git directories. ```python def select_device(device='', batch_size=None): # device = 'cpu' or '0' or '0,1,2,3' s = f'YOLOv5 🚀 {git_describe() or date_modified()} torch {torch.__version__} ' # string ... ``` * Update detections() self.t = tuple() (ultralytics#2617) * Update detections() self.t = tuple() Fix multiple results.print() bug. * Update experimental.py * Update yolo.py * Fix Indentation in test.py (ultralytics#2614) * Fix Indentation in test.py * CI fix * Comply with PEP8: 80 characters per line * Update Detections() self.n comment (ultralytics#2620) ```python self.n = len(self.pred) # number of images (batch size) ``` * Remove conflicting nvidia-tensorboard package (ultralytics#2622) Attempt to resolve tensorboard Docker error in ultralytics#2573 * FROM nvcr.io/nvidia/pytorch:21.03-py3 (ultralytics#2623) Update Docker FROM nvcr.io/nvidia/pytorch:21.03-py3 * Improve git_describe() (ultralytics#2633) Catch 'fatal: not a git repository' returns and return '' instead (observed in GCP Hub checks). * Fix: evolve with wandb (ultralytics#2634) * W&B resume ddp from run link fix (ultralytics#2579) * W&B resume ddp from run link fix * Native DDP W&B support for training, resuming * Improve git_describe() fix 1 (ultralytics#2635) Add stderr=subprocess.STDOUT to catch error messages. * PyTorch Hub custom model to CUDA device fix (ultralytics#2636) Fix for ultralytics#2630 raised by @Pro100rus32 * PyTorch Hub amp.autocast() inference (ultralytics#2641) I think this should help speed up CUDA inference, as currently models may be running in FP32 inference mode on CUDA devices unnecesarily. * Add tqdm pbar.close() (ultralytics#2644) When using tqdm, sometimes it can't print in one line and roll to next line. * Speed profiling improvements (ultralytics#2648) * Speed profiling improvements * Update torch_utils.py deepcopy() required to avoid adding elements to model. * Update torch_utils.py * Created using Colaboratory (ultralytics#2649) * Update requirements.txt (ultralytics#2564) * Add opencv-contrib-python to requirements.txt * Update requirements.txt Co-authored-by: Glenn Jocher <[email protected]> * add option to disable half precision in test.py (ultralytics#2507) Co-authored-by: Glenn Jocher <[email protected]> * Add --label-smoothing eps argument to train.py (default 0.0) (ultralytics#2344) * Add label smoothing option * Correct data type * add_log * Remove log * Add log * Update loss.py remove comment (too versbose) Co-authored-by: phattran <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> * Created using Colaboratory * Set resume flag to false (ultralytics#2657) * Update README.md * Created using Colaboratory * Update README with Tips for Best Results tutorial (ultralytics#2682) * Update README with Tips for Best Results tutorial * Update README.md * Add TransformerLayer, TransformerBlock, C3TR modules (ultralytics#2333) * yolotr * transformer block * Remove bias in Transformer * Remove C3T * Remove a deprecated class * put the 2nd LayerNorm into the 2nd residual block * move example model to models/hub, rename to -transformer * Add module comments and TODOs * Remove LN in Transformer * Add comments for Transformer * Solve the problem of MA with DDP * cleanup * cleanup find_unused_parameters * PEP8 reformat Co-authored-by: DingYiwei <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> * Fix: ultralytics#2674 (ultralytics#2683) * Set resume flag to false * Check existance of val dataset * PyTorch Hub model.save() increment as runs/hub/exp (ultralytics#2684) * PyTorch Hub model.save() increment as runs/hub/exp This chane will align PyTorch Hub results saving with the existing unified results saving directory structure of runs/ /train /detect /test /hub /exp /exp2 ... * cleanup * autoShape forward im = np.asarray(im) # to numpy (ultralytics#2689) Slight speedup. * pip install coremltools onnx (ultralytics#2690) Requested in ultralytics#2686 * Updated filename attributes for YOLOv5 Hub results (ultralytics#2708) Proposed fix for 'Model predict with forward will fail if PIL image does not have filename attribute' ultralytics#2702 * Updated filename attributes for YOLOv5 Hub BytesIO (ultralytics#2718) Fix 2 for 'Model predict with forward will fail if PIL image does not have filename attribute' ultralytics#2702 * Add support for list-of-directory data format for wandb (ultralytics#2719) * Update README with collapsable notes (ultralytics#2721) * Update README with collapsable notes. * cleanup * center table * Add Hub results.pandas() method (ultralytics#2725) * Add Hub results.pandas() method New method converts results from torch tensors to pandas DataFrames with column names. This PR may partially resolve issue ultralytics#2703 ```python results = model(imgs) print(results.pandas().xyxy[0]) xmin ymin xmax ymax confidence class name 0 57.068970 391.770599 241.383545 905.797852 0.868964 0 person 1 667.661255 399.303589 810.000000 881.396667 0.851888 0 person 2 222.878387 414.774231 343.804474 857.825073 0.838376 0 person 3 4.205386 234.447678 803.739136 750.023376 0.658006 5 bus 4 0.000000 550.596008 76.681190 878.669922 0.450596 0 person ``` * Update comments torch example input now shown resized to size=640 and also now a multiple of P6 stride 64 (see ultralytics#2722 (comment)) * apply decorators * PEP8 * Update common.py * pd.options.display.max_columns = 10 * Update common.py * autocast enable=torch.cuda.is_available() (ultralytics#2748) * torch.cuda.amp bug fix (ultralytics#2750) PR ultralytics#2725 introduced a very specific bug that only affects multi-GPU trainings. Apparently the cause was using the torch.cuda.amp decorator in the autoShape forward method. I've implemented amp more traditionally in this PR, and the bug is resolved. * utils/wandb_logging PEP8 reformat (ultralytics#2755) * wandb_logging PEP8 reformat * Update wandb_utils.py * Tensorboard model visualization bug fix (ultralytics#2758) This fix should allow for visualizing YOLOv5 model graphs correctly in Tensorboard by uncommenting line 335 in train.py: ```python if tb_writer: tb_writer.add_graph(torch.jit.trace(model, imgs, strict=False), []) # add model graph ``` The problem was that the detect() layer checks the input size to adapt the grid if required, and tracing does not seem to like this shape check (even if the shape is fine and no grid recomputation is required). The following will warn: https://github.com/ultralytics/yolov5/blob/0cae7576a9241110157cd154fc2237e703c2719e/train.py#L335 Solution is below. This is a YOLOv5s model displayed in TensorBoard. You can see the Detect() layer merging the 3 layers into a single output for example, and everything appears to work and visualize correctly. ```python tb_writer.add_graph(torch.jit.trace(model, imgs, strict=False), []) ``` <img width="893" alt="Screenshot 2021-04-11 at 01 10 09" src="https://user-images.githubusercontent.com/26833433/114286928-349bd600-9a63-11eb-941f-7139ee6cd602.png"> * Created using Colaboratory * YouTube Livestream Detection (ultralytics#2752) * Youtube livestream detection * dependancy update to auto install pafy * Remove print * include youtube_dl in deps * PEP8 reformat * youtube url check fix * reduce lines * add comment * update check_requirements * stream framerate fix * Update README.md * cleanup * PEP8 * remove cap.retrieve() failure code Co-authored-by: Glenn Jocher <[email protected]> * YOLOv5 v5.0 Release (ultralytics#2762) * YOLOv5 v5.0 Release patch 1 (ultralytics#2764) * torch.jit.trace(model, img, strict=False) * Update check_file() * Update hubconf.py * Update README.md * Update tutorial.ipynb * Created using Colaboratory * Update tutorial.ipynb * Created using Colaboratory * Created using Colaboratory * Update README.md * Flask REST API Example (ultralytics#2732) * add files * Update README.md * Update README.md * Update restapi.py pretrained=True and model.eval() are used by default when loading a model now, so no need to call them manually. * PEP8 reformat * PEP8 reformat Co-authored-by: Glenn Jocher <[email protected]> * Update README.md * ONNX Simplifier (ultralytics#2815) * ONNX Simplifier Add ONNX Simplifier to ONNX export pipeline in export.py. Will auto-install onnx-simplifier if onnx is installed but onnx-simplifier is not. * Update general.py * YouTube Bug Fix (ultralytics#2818) Fix for ultralytics#2810 ```shell python detect.py --source 0 ``` introduced by YouTube Livestream Detection PR ultralytics#2752 * PyTorch Hub cv2 .save() .show() bug fix (ultralytics#2831) * PyTorch Hub cv2 .save() .show() bug fix cv2.rectangle() was failing on non-contiguous np array inputs. This checks for contiguous arrays and applies is necessary: ```python imgs[i] = im if im.data.contiguous else np.ascontiguousarray(im) # update ``` * Update plots.py ```python assert im.data.contiguous, 'Image not contiguous. Apply np.ascontiguousarray(im) to plot_on_box() input image.' ``` * Update hubconf.py Expand CI tests to OpenCV image. * Create FUNDING.yml (ultralytics#2832) * Update FUNDING.yml (ultralytics#2833) * Update FUNDING.yml * move FUNDING.yml to ./github * Fix ONNX dynamic axes export support with onnx simplifier, make onnx simplifier optional (ultralytics#2856) * Ensure dynamic export works succesfully, onnx simplifier optional * Update export.py * add dashes Co-authored-by: Tim <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> * Update increment_path() to handle file paths (ultralytics#2867) * Detection cropping+saving feature addition for detect.py and PyTorch Hub (ultralytics#2827) * Update detect.py * Update detect.py * Update greetings.yml * Update cropping * cleanup * Update increment_path() * Update common.py * Update detect.py * Update detect.py * Update detect.py * Update common.py * cleanup * Update detect.py Co-authored-by: Glenn Jocher <[email protected]> * Implement yaml.safe_load() (ultralytics#2876) * Implement yaml.safe_load() * yaml.safe_dump() * Cleanup load_image() (ultralytics#2871) * don't resize up in load_image if augmenting * cleanup Co-authored-by: Glenn Jocher <[email protected]> * bug fix: switched rows and cols for correct detections in confusion matrix (ultralytics#2883) * VisDrone2019-DET Dataset Auto-Download (ultralytics#2882) * VisDrone Dataset Auto-Download * add visdrone.yaml * cleanup * add VisDrone2019-DET-test-dev * cleanup VOC * Uppercase model filenames enabled (ultralytics#2890) * ACON activation function (ultralytics#2893) * ACON Activation Function ## 🚀 Feature There is a new activation function [ACON (CVPR 2021)](https://arxiv.org/pdf/2009.04759.pdf) that unifies ReLU and Swish. ACON is simple but very effective, code is here: https://github.com/nmaac/acon/blob/main/acon.py#L19 ![image](https://user-images.githubusercontent.com/5032208/115676962-a38dfe80-a382-11eb-9883-61fa3216e3e6.png) The improvements are very significant: ![image](https://user-images.githubusercontent.com/5032208/115680180-eac9be80-a385-11eb-9c7a-8643db552c69.png) ## Alternatives It also has an enhanced version meta-ACON that uses a small network to learn beta explicitly, which may influence the speed a bit. ## Additional context [Code](https://github.com/nmaac/acon) and [paper](https://arxiv.org/pdf/2009.04759.pdf). * Update activations.py * Explicit opt function arguments (ultralytics#2817) * more explicit function arguments * fix typo in detect.py * revert import order * revert import order * remove default value * Update yolo.py (ultralytics#2899) * Update google_utils.py (ultralytics#2900) * Add detect.py --hide-conf --hide-labels --line-thickness options (ultralytics#2658) * command line option for line thickness and hiding labels * command line option for line thickness and hiding labels * command line option for line thickness and hiding labels * command line option for line thickness and hiding labels * command line option for line thickness and hiding labels * command line option for hiding confidence values * Update detect.py Co-authored-by: Glenn Jocher <[email protected]> * Default optimize_for_mobile() on TorchScript models (ultralytics#2908) Per https://pytorch.org/tutorials/recipes/script_optimized.html this should improve performance on torchscript models (and maybe coreml models also since coremltools operates on a torchscript model input, though this still requires testing). * Update export.py (ultralytics#2909) * Update export.py for 2 dry runs (ultralytics#2910) * Update export.py for 2 dry runs * Update export.py * Add file_size() function (ultralytics#2911) * Add file_size() function * Update export.py * Update download() for tar.gz files (ultralytics#2919) * Update download() for tar.gz files * Update general.py * Update visdrone.yaml (ultralytics#2921) * Change default value of hide label argument to False (ultralytics#2923) * Change default value of hide-conf argument to false (ultralytics#2925) * test.py native --single-cls (ultralytics#2928) * Add verbose option to pytorch hub models (ultralytics#2926) * Add verbose and update print to logging * Fix positonal param * Revert auto formatting changes * Update hubconf.py Co-authored-by: Glenn Jocher <[email protected]> * ACON Activation batch-size 1 bug patch (ultralytics#2901) * ACON Activation batch-size 1 bug path This is not a great solution to nmaac/acon#4 but it's all I could think of at the moment. WARNING: YOLOv5 models with MetaAconC() activations are incapable of running inference at batch-size 1 properly due to a known bug in nmaac/acon#4 with no known solution. * Update activations.py * Update activations.py * Update activations.py * Update activations.py * Check_requirements() enclosing apostrophe bug fix (ultralytics#2929) This fixes a bug where the '>' symbol in python package requirements was not running correctly with subprocess.check_output() commands. * Update README.md (ultralytics#2934) * Update README.md dependencies: ImportError: libGL.so.1: cannot open shared object file: No such file or directory ImportError: libgthread-2.0.so.0: cannot open shared object file: No such file or directory ImportError: libSM.so.6: cannot open shared object file: No such file or directory ImportError: libXrender.so.1: cannot open shared object file: No such file or directory * replace older apt-get with apt Code commented for now until a better understanding of the issue, and also code is not cross-platform compatible. Co-authored-by: Glenn Jocher <[email protected]> * Improved yolo.py profiling (ultralytics#2940) * Improved yolo.py profiling Improved column order and labelling. * Update yolo.py * Add yolov5/ to sys.path() for *.py subdir exec (ultralytics#2949) * Add yolov5/ to sys.path() for *.py subdir exec * Update export.py * update UI + latest yolov5 sources (#15) * merge latest version done, not tested * split tabs with radio buttons * models table -wip * models table -wip * start split html template to parts * ui refactoring * compile-template wip - paths confusion * compile wip * train/val splits * keep/ignore unlabeled images * models table * training hyperparameters * UI templates - done * unlabeled count in UI * add adam optimizer * convert_project to detection - works * start train/val splits * splits wip * splits done, only simple tests * splits validation * data preprocessing - not tested * download weights - wip * init_script_arguments - not tested * init_script_arguments - not tested * prepare weights - wip * not tested * add metrics period * set output * artifacts dirs * train_batches_uploaded flag * pre-release for debug * update config * update SDK version * fix imports * change imports * change imports * add UI sources directory to sys.path * new SDK version * new SDK version * fix GIoU smoothing * update smoothing * send metrics for the last epoch * save link to app UI * todo * log train/val size * sly-to-yolov5 format: fix same names in different datasets * fix inference * serve not tested * [serve] modal table stat * [serve] modal tabs * [serve] modal tabs * [serve] modal width * [serve] modal tabs style * [serve] fix pretrained weights URL * [serve] add stride to serv * [train] readme wip * [train] readme wip * [train] readme wip * [serve] change inference_image_id to work with remote storages (s3, azure, ...) * [serve] fix stride initialization * [serve] yolov5 serve - fixed * add additional info logs * [serve] todo * [train] splits - hide notice1 * fix collections readme * train readme - new screenshot * train readme Co-authored-by: Abhiram V <[email protected]> Co-authored-by: Glenn Jocher <[email protected]> Co-authored-by: ramonhollands <[email protected]> Co-authored-by: Ayush Chaurasia <[email protected]> Co-authored-by: train255 <[email protected]> Co-authored-by: ab-101 <[email protected]> Co-authored-by: Transigent <[email protected]> Co-authored-by: NanoCode012 <[email protected]> Co-authored-by: Daniel Khromov <[email protected]> Co-authored-by: VdLMV <[email protected]> Co-authored-by: Jaap van de Loosdrecht <[email protected]> Co-authored-by: Yann Defretin <[email protected]> Co-authored-by: Aditya Lohia <[email protected]> Co-authored-by: Shivam Swanrkar <[email protected]> Co-authored-by: xiaowo1996 <[email protected]> Co-authored-by: Iden Craven <[email protected]> Co-authored-by: Jan Hajek <[email protected]> Co-authored-by: Jan Hajek <[email protected]> Co-authored-by: oleg <[email protected]> Co-authored-by: Oleg Boiko <[email protected]> Co-authored-by: Ryan Avery <[email protected]> Co-authored-by: Yonghye Kwon <[email protected]> Co-authored-by: Kartikeya Sharma <[email protected]> Co-authored-by: Kartikeya Sharma <[email protected]> Co-authored-by: Kartikeya Sharma <[email protected]> Co-authored-by: Yann Defretin <[email protected]> Co-authored-by: maxupp <[email protected]> Co-authored-by: Max Uppenkamp <[email protected]> Co-authored-by: zzttqu <[email protected]> Co-authored-by: Youngjin Shin <[email protected]> Co-authored-by: Benjamin Fineran <[email protected]> Co-authored-by: Phat Tran <[email protected]> Co-authored-by: phattran <[email protected]> Co-authored-by: Ding Yiwei <[email protected]> Co-authored-by: DingYiwei <[email protected]> Co-authored-by: Ben Milanko <[email protected]> Co-authored-by: Robin <[email protected]> Co-authored-by: Tim Stokman <[email protected]> Co-authored-by: Tim <[email protected]> Co-authored-by: Burhan <[email protected]> Co-authored-by: JoshSong <[email protected]> Co-authored-by: Michael Heilig <[email protected]> Co-authored-by: r-blmnr <[email protected]> Co-authored-by: fcakyon <[email protected]> Co-authored-by: Maximilian Peters <[email protected]> Co-authored-by: albinxavi <[email protected]> Co-authored-by: BZFYS <[email protected]>

(cherry picked from commit f5b8f7d)

YOLOv5 v5.0 Release

2f214bd

glenn-jocher added the enhancement New feature or request label Apr 11, 2021

glenn-jocher merged commit f5b8f7d into master Apr 11, 2021

glenn-jocher deleted the v5 branch April 11, 2021 17:23

glenn-jocher mentioned this pull request Apr 11, 2021

YOLOv5 v5.0 Release patch 1 #2764

Merged

glenn-jocher self-assigned this Apr 11, 2021

KMint1819 pushed a commit to KMint1819/yolov5 that referenced this pull request May 12, 2021

YOLOv5 v5.0 Release (ultralytics#2762)

75c6e5d

Lechtr pushed a commit to Lechtr/yolov5 that referenced this pull request Jul 20, 2021

YOLOv5 v5.0 Release (ultralytics#2762)

7edf5bf

(cherry picked from commit f5b8f7d)

taicaile pushed a commit to taicaile/yolov5 that referenced this pull request Oct 12, 2021

YOLOv5 v5.0 Release (ultralytics#2762)

1fa7a20

zhiqwang mentioned this pull request Oct 16, 2021

Support upstream yolov5 P6 models zhiqwang/yolort#196

Merged

BjarneKuehl pushed a commit to fhkiel-mlaip/yolov5 that referenced this pull request Aug 26, 2022

YOLOv5 v5.0 Release (ultralytics#2762)

36697f9

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

YOLOv5 v5.0 Release #2762

YOLOv5 v5.0 Release #2762

glenn-jocher commented Apr 11, 2021 •

edited by UltralyticsAssistant

Loading

AyushExel commented Apr 11, 2021

Edwardmark commented Apr 16, 2021

glenn-jocher commented Apr 16, 2021 •

edited

Loading

YOLOv5 v5.0 Release #2762

YOLOv5 v5.0 Release #2762

Conversation

glenn-jocher commented Apr 11, 2021 • edited by UltralyticsAssistant Loading

Notable Updates

Updated Results

Pretrained Checkpoints

Changelog

🛠️ PR Summary

🌟 Summary

📊 Key Changes

🎯 Purpose & Impact

AyushExel commented Apr 11, 2021

Edwardmark commented Apr 16, 2021

glenn-jocher commented Apr 16, 2021 • edited Loading

glenn-jocher commented Apr 11, 2021 •

edited by UltralyticsAssistant

Loading

glenn-jocher commented Apr 16, 2021 •

edited

Loading