v6.2 - YOLOv5 Classification Models, Apple M1, Reproducibility, ClearML and Deci.ai integrations

[glenn-jocher]

This release incorporates 401 PRs from 41 contributors since our last release in February 2022. It adds Classification training, validation, prediction and export (to all 11 formats), and also provides ImageNet-pretrained YOLOv5m-cls, ResNet (18, 34, 50, 101) and EfficientNet (b0-b3) models.

My main goal with this release is to introduce super simple YOLOv5 classification workflows just like our existing object detection models. The new v6.2 YOLOv5-cls models below are just a start, we will continue to improve these going forward together with our existing detection models. We'd love your contributions on this effort!

Our next release, v6.3 is scheduled for September and will bring official instance segmentation support to YOLOv5, with a major v7.0 release later this year updating architectures across all 3 tasks - classification, detection and segmentation.

Important Updates

• Classification Models ⭐ NEW: YOLOv5-cls ImageNet-pretrained classification models are now available for the first time (New YOLOv5 Classification Models #8956 by @glenn-jocher)
• ClearML logging ⭐ NEW: Integration with the open-source experiment tracker ClearML. Installing with pip install clearml will enable the integration and allow users to track every training run in ClearML. This in turn allows users to track and compare runs and even schedule runs remotely. (ClearML experiment tracking integration #8620 by @thepycoder)
• Deci.ai optimization ⭐ NEW: Automatically compile and quantize YOLOv5 for better inference performance in one click at Deci (New YOLOv5 Classification Models #8956 by @glenn-jocher).
• GPU Export Benchmarks: Benchmark (mAP and speed) all YOLOv5 export formats with python utils/benchmarks.py --weights yolov5s.pt --device 0 for GPU benchmarks or --device cpu for CPU benchmarks (YOLOv5 Export Benchmarks for GPU #6963 by @glenn-jocher).
• Training Reproducibility: Single-GPU YOLOv5 training with torch>=1.12.0 is now fully reproducible, and a new --seed argument can be used (default seed=0) (Training reproducibility improvements #8213 by @AyushExel).
• Apple Metal Performance Shader (MPS) Support: MPS support for Apple M1/M2 devices with --device mps (full functionality is pending torch updates in General MPS op coverage tracking issue pytorch/pytorch#77764) (YOLOv5 Apple Metal Performance Shader (MPS) support #7878 by @glenn-jocher)

New Classification Checkpoints

We trained YOLOv5-cls classification models on ImageNet for 90 epochs using a 4xA100 instance, and we trained ResNet and EfficientNet models alongside with the same default training settings to compare. We exported all models to ONNX FP32 for CPU speed tests and to TensorRT FP16 for GPU speed tests. We ran all speed tests on Google Colab Pro for easy reproducibility.

Model	size (pixels)	accuracy top1	accuracy top5	Train time 90 epochs 4x A100 (hours)	Speed ONNX-CPU (ms)	Speed TensorRT-V100 (ms)	params (M)	FLOPs @224 (B)
YOLOv5n-cls	224	64.6	85.4	7:59	3.3	0.5	2.5	0.5
YOLOv5s-cls	224	71.5	90.2	8:09	6.6	0.6	5.4	1.4
YOLOv5m-cls	224	75.9	92.9	10:06	15.5	0.9	12.9	3.9
YOLOv5l-cls	224	78.0	94.0	11:56	26.9	1.4	26.5	8.5
YOLOv5x-cls	224	79.0	94.4	15:04	54.3	1.8	48.1	15.9
ResNet18	224	70.3	89.5	6:47	11.2	0.5	11.7	3.7
ResNet34	224	73.9	91.8	8:33	20.6	0.9	21.8	7.4
ResNet50	224	76.8	93.4	11:10	23.4	1.0	25.6	8.5
ResNet101	224	78.5	94.3	17:10	42.1	1.9	44.5	15.9
EfficientNet_b0	224	75.1	92.4	13:03	12.5	1.3	5.3	1.0
EfficientNet_b1	224	76.4	93.2	17:04	14.9	1.6	7.8	1.5
EfficientNet_b2	224	76.6	93.4	17:10	15.9	1.6	9.1	1.7
EfficientNet_b3	224	77.7	94.0	19:19	18.9	1.9	12.2	2.4

• All checkpoints are trained to 90 epochs with SGD optimizer with lr0=0.001 and weight_decay=5e-5 at image size 224 and all default settings.
  Runs logged to https://wandb.ai/glenn-jocher/YOLOv5-Classifier-v6-2
• Accuracy values are for single-model single-scale on ImageNet-1k dataset.
  Reproduce by python classify/val.py --data ../datasets/imagenet --img 224
• Speed averaged over 100 inference images using a Colab Pro V100 High-RAM instance.
  Reproduce by python classify/val.py --data ../datasets/imagenet --img 224 --batch 1
• Export to ONNX at FP32 and TensorRT at FP16 done with export.py.
  Reproduce by python export.py --weights yolov5s-cls.pt --include engine onnx --imgsz 224

New Classification Model Usage Examples

Train

YOLOv5 classification training supports auto-download of MNIST, Fashion-MNIST, CIFAR10, CIFAR100, Imagenette, Imagewoof, and ImageNet datasets with the --data argument. To start training on MNIST for example use --data mnist.

# Single-GPU
python classify/train.py --model yolov5s-cls.pt --data cifar100 --epochs 5 --img 224 --batch 128

# Multi-GPU DDP
python -m torch.distributed.run --nproc_per_node 4 --master_port 1 classify/train.py --model yolov5s-cls.pt --data imagenet --epochs 5 --img 224 --device 0,1,2,3

Val

Validate YOLOv5m-cls accuracy on ImageNet-1k dataset:

bash data/scripts/get_imagenet.sh --val  # download ImageNet val split (6.3G, 50000 images)
python classify/val.py --weights yolov5m-cls.pt --data ../datasets/imagenet --img 224  # validate

Predict

Use pretrained YOLOv5s-cls.pt to predict bus.jpg:

python classify/predict.py --weights yolov5s-cls.pt --data data/images/bus.jpg

model = torch.hub.load('ultralytics/yolov5', 'custom', 'yolov5s-cls.pt')  # load from PyTorch Hub

Export

Export a group of trained YOLOv5s-cls, ResNet and EfficientNet models to ONNX and TensorRT:

python export.py --weights yolov5s-cls.pt resnet50.pt efficientnet_b0.pt --include onnx engine --img 224

Changelog

• Changes between previous release and this release: v6.1...v6.2
• Changes since this release: v6.2...HEAD

🛠️ New Features and Bug Fixes (401)

• Pre-commit table fix by @glenn-jocher in Pre-commit table fix #6744
• Update tutorial.ipynb (2 CPUs, 12.7 GB RAM, 42.2/166.8 GB disk) by @glenn-jocher in Update tutorial.ipynb (2 CPUs, 12.7 GB RAM, 42.2/166.8 GB disk) #6767
• Update min warmup iterations from 1k to 100 by @glenn-jocher in Update min warmup iterations from 1k to 100 #6768
• Default OMP_NUM_THREADS=8 by @glenn-jocher in Default OMP_NUM_THREADS=8 #6770
• Update tutorial.ipynb by @glenn-jocher in Update tutorial.ipynb #6771
• Update hyp.VOC.yaml by @glenn-jocher in Update hyp.VOC.yaml #6772
• Fix export for 1-channel images by @lcombaldieu in Fix export for 1-channel images #6780
• Update EMA decay tau by @glenn-jocher in Update EMA decay tau #6769
• YOLOv5s6 params FLOPs fix by @glenn-jocher in YOLOv5s6 params FLOPs fix #6782
• Update PULL_REQUEST_TEMPLATE.md by @glenn-jocher in Update PULL_REQUEST_TEMPLATE.md #6783
• Update autoanchor.py by @glenn-jocher in Update autoanchor.py #6794
• Update sweep.yaml by @lcombaldieu in Update sweep.yaml #6825
• AutoAnchor improved initialization robustness by @glenn-jocher in AutoAnchor improved initialization robustness #6854
• Add *.ts to VID_FORMATS by @glenn-jocher in Add *.ts to VID_FORMATS #6859
• Update --cache disk deprecate *_npy/ dirs by @glenn-jocher in Update --cache disk deprecate *_npy/ dirs #6876
• Update yolov5s.yaml by @vnekat in Update yolov5s.yaml #6865
• Default FP16 TensorRT export by @DavidBaldsiefen in Default FP16 TensorRT export #6798
• Bump actions/setup-python from 2 to 3 by @dependabot in Bump actions/setup-python from 2 to 3 #6880
• Bump actions/checkout from 2 to 3 by @dependabot in Bump actions/checkout from 2 to 3 #6881
• Fix TRT max_workspace_size deprecation notice by @glenn-jocher in Fix TRT max_workspace_size deprecation notice #6856
• Update bytes to GB with bit shift by @glenn-jocher in Update bytes to GB with bit shift #6886
• Move git_describe() to general.py by @glenn-jocher in Move git_describe() to general.py #6918
• PyTorch 1.11.0 compatibility updates by @glenn-jocher in PyTorch 1.11.0 compatibility updates #6932
• Optimize PyTorch 1.11.0 compatibility update by @glenn-jocher in Optimize PyTorch 1.11.0 compatibility update #6933
• Allow 3-point segments by @glenn-jocher in Allow 3-point segments #6938
• Fix PyTorch Hub export inference shapes by @glenn-jocher in Fix PyTorch Hub export inference shapes #6949
• DetectMultiBackend() --half handling by @glenn-jocher in DetectMultiBackend() --half handling #6945
• Update Dockerfile torch==1.11.0+cu113 by @glenn-jocher in Update Dockerfile torch==1.11.0+cu113 #6954
• New val.py cuda variable by @glenn-jocher in New val.py cuda variable #6957
• DetectMultiBackend() return device update by @glenn-jocher in DetectMultiBackend() return device update #6958
• Tensor initialization on device improvements by @glenn-jocher in Tensor initialization on device improvements #6959
• EdgeTPU optimizations by @paradigmn in EdgeTPU optimizations #6808
• Model ema key backward compatibility fix by @glenn-jocher in Model ema key backward compatibility fix #6972
• YOLOv5 Export Benchmarks for GPU by @glenn-jocher in YOLOv5 Export Benchmarks for GPU #6963
• Update TQDM bar format by @glenn-jocher in Update TQDM bar format #6988
• Conditional Timeout() by OS (disable on Windows) by @glenn-jocher in Conditional Timeout() by OS (disable on Windows) #7013
• fix: add default PIL font as fallback by @maxstrobel in fix: add default PIL font as fallback  #7010
• Consistent saved_model output format by @MrinalJain17 in Consistent saved_model output format #7032
• ComputeLoss() indexing/speed improvements by @glenn-jocher in ComputeLoss() indexing/speed improvements #7048
• Update Dockerfile to git clone instead of COPY by @glenn-jocher in Update Dockerfile to git clone instead of COPY #7053
• Create SECURITY.md by @glenn-jocher in Create SECURITY.md #7054
• Fix incomplete URL substring sanitation by @glenn-jocher in Fix incomplete URL substring sanitation #7056
• Use PIL to eliminate chroma subsampling in crops by @LaserBorg in Use PIL to eliminate chroma subsampling in crops #7008
• Fix check_anchor_order() in pixel-space not grid-space by @glenn-jocher in Fix check_anchor_order() in pixel-space not grid-space #7060
• Update detect.py non-inplace with y.tensor_split() by @glenn-jocher in Update detect.py non-inplace with y.tensor_split() #7062
• Update common.py lists for tuples by @glenn-jocher in Update common.py lists for tuples #7063
• Update W&B message to LOGGER.info() by @glenn-jocher in Update W&B message to LOGGER.info() #7064
• Update init.py by @glenn-jocher in Update __init__.py #7065
• Add non-zero da check_anchor_order() condition by @glenn-jocher in Add non-zero da check_anchor_order() condition #7066
• Fix2 check_anchor_order() in pixel-space not grid-space by @glenn-jocher in Fix2 check_anchor_order() in pixel-space not grid-space #7067
• Revert "Update detect.py non-inplace with y.tensor_split()" by @glenn-jocher in Revert "Update detect.py non-inplace with y.tensor_split()" #7074
• Update loss.py with if self.gr < 1: by @glenn-jocher in Update loss.py with if self.gr < 1: #7087
• Update loss for FP16 tobj by @glenn-jocher in Update loss for FP16 tobj #7088
• Display model name in model summary by @glenn-jocher in Display model name in model summary #7101
• torch.split() 1.7.0 compatibility fix by @glenn-jocher in torch.split() 1.7.0 compatibility fix #7102
• Update benchmarks significant digits by @glenn-jocher in Update benchmarks significant digits #7103
• Model summary pathlib fix by @glenn-jocher in Model summary pathlib fix #7104
• Remove named arguments where possible by @glenn-jocher in Remove named arguments where possible #7105
• Multi-threaded VisDrone and VOC downloads by @glenn-jocher in Multi-threaded VisDrone and VOC downloads #7108
• np.fromfile() Chinese image paths fix by @yeshanliu in np.fromfile() Chinese image paths fix #6979
• Add PyTorch Hub results.save(labels=False) option by @glenn-jocher in Add PyTorch Hub results.save(labels=False) option #7129
• Fix cv2.imwrite on non-ASCII paths by @CCRcmcpe in Fix cv2.imwrite on non-ASCII paths #7139
• Fix detect.py --view-img for non-ASCII paths by @Zengyf-CVer in Fix detect.py --view-img for non-ASCII paths #7093
• Add Architecture Summary to README Tutorials by @glenn-jocher in Add Architecture Summary to README Tutorials #7146
• Adjust NMS time limit warning to batch size by @glenn-jocher in Adjust NMS time limit warning to batch size #7156
• Sidestep os.path.relpath() Windows bug by @glenn-jocher in Sidestep os.path.relpath() Windows bug #7158
• NMS unused variable fix by @glenn-jocher in NMS unused variable fix #7161
• yolo.py --profile updates by @glenn-jocher in yolo.py --profile updates #7170
• Revert C3() change by @glenn-jocher in Revert C3() change #7172
• Bump actions/cache from 2.1.7 to 3 by @dependabot in Bump actions/cache from 2.1.7 to 3 #7175
• yolo.py profiling updates by @glenn-jocher in yolo.py profiling updates #7178
• Update tutorial.ipynb by @glenn-jocher in Update tutorial.ipynb #7212
• ENV OMP_NUM_THREADS=8 by @glenn-jocher in ENV OMP_NUM_THREADS=8 #7215
• Add train.py --name cfg option by @glenn-jocher in Add train.py --name cfg option #7202
• precommit: yapf by @Borda in precommit: yapf #5494
• CLI fire prep updates by @glenn-jocher in CLI fire prep updates #7229
• Update .pre-commit-config.yaml by @glenn-jocher in Update .pre-commit-config.yaml #7230
• SavedModel TF Serve Fix by @glenn-jocher in SavedModel TF Serve Fix #7228
• Create CODE_OF_CONDUCT.md by @glenn-jocher in Create CODE_OF_CONDUCT.md #7233
• Fix www.youtube.com hostname by @glenn-jocher in Fix www.youtube.com hostname #7242
• Update minimum Python>=3.7.0 by @glenn-jocher in Update minimum Python>=3.7.0 #7247
• Update setup.cfg to description_file field by @glenn-jocher in Update setup.cfg to description_file field #7248
• Update tutorial.ipynb by @glenn-jocher in Update tutorial.ipynb #7254
• Update tutorial.ipynb by @glenn-jocher in Update tutorial.ipynb #7255
• Fix Flask REST API by @Zengyf-CVer in Fix Flask REST API #7210
• Export with official nn.SiLU() by @glenn-jocher in Export with official nn.SiLU() #7256
• Refactor out-of-place Detect() for reduced ops by @glenn-jocher in Refactor out-of-place Detect() for reduced ops #7257
• torch.split() replace slicing on out-of-place inference by @glenn-jocher in torch.split() replace slicing on out-of-place inference #7258
• Export single output only by @glenn-jocher in Export single output only #7259
• TorchScript single-output fix by @glenn-jocher in TorchScript single-output fix #7261
• Integrate offset into grid by @glenn-jocher in Integrate offset into grid #7262
• [pre-commit.ci] pre-commit suggestions by @pre-commit-ci in [pre-commit.ci] pre-commit suggestions #7279
• Update Dockerfile by @glenn-jocher in Update Dockerfile #7282
• Enable TensorFlow ops for --nms and --agnostic-nms by @leeflix in Enable TensorFlow ops for --nms and --agnostic-nms #7281
• Update cv2.imread() patch with flags argument by @glenn-jocher in Update cv2.imread() patch with flags argument #7287
• Context manager open(file) as f fixes by @glenn-jocher in Context manager open(file) as f fixes #7289
• val.py --weights and --data compatibility check by @glenn-jocher in val.py --weights and --data compatibility check #7292
• Add dataset sizes (zipped) by @glenn-jocher in Add dataset sizes (zipped) #7293
• Add check_requirements(('pycocotools>=2.0',)) by @glenn-jocher in Add check_requirements(('pycocotools>=2.0',)) #7295
• fix: disable usage of root logger by @maxstrobel in fix: disable usage of root logger #7296
• Update export.py by @glenn-jocher in Update export.py #7301
• Use tqdm.auto by @glenn-jocher in Use tqdm.auto #7311
• Add retry=3 to download() by @glenn-jocher in Add retry=3 to download() #7313
• Add callbacks by @glenn-jocher in Add callbacks #7315
• Copy wandb param dict before training to avoid overwrites. by @n1mmy in Copy wandb param dict before training to avoid overwrites. #7317
• Update Objects365.yaml by @glenn-jocher in Update Objects365.yaml #7323
• Fix TF export for BottleneckCSP layer by @nrupatunga in Fix TF export for BottleneckCSP layer #7330
• Objects365 images GB vs zips GB by @glenn-jocher in Objects365 images GB vs zips GB #7335
• Simplify callbacks.py return by @glenn-jocher in Simplify callbacks.py return #7333
• Print dataset scan only if RANK in (-1, 0) by @glenn-jocher in Print dataset scan only if RANK in (-1, 0) #7337
• Update _make_grid() to model dtype by @glenn-jocher in Update _make_grid() to model dtype #7346
• Rename 'MacOS' to 'macOS' by @glenn-jocher in Rename 'MacOS' to 'macOS' #7349
• Add python benchmarks.py --test for export-only by @glenn-jocher in Add python benchmarks.py --test for export-only #7350
• Add ONNX export metadata by @glenn-jocher in Add ONNX export metadata #7353
• DetectMultiBackend() default stride=32 by @rglkt in DetectMultiBackend() default stride=32 #7342
• Loss and IoU speed improvements by @glenn-jocher in Loss and IoU speed improvements #7361
• Swap unsafe_chunk() for chunk() by @glenn-jocher in Swap unsafe_chunk() for chunk() #7362
• Delete FUNDING.yml by @glenn-jocher in Delete FUNDING.yml #7363
• Replace Slack with Community Forum in issues by @glenn-jocher in Replace Slack with Community Forum in issues #7364
• Update ci-testing.yml by @glenn-jocher in Update ci-testing.yml #7365
• Bump actions/stale from 4 to 5 by @dependabot in Bump actions/stale from 4 to 5 #7371
• Update optimizer param group strategy by @glenn-jocher in Update optimizer param group strategy #7376
• Add support for different normalization layers by @vardanagarwal in Add support for different normalization layers #7377
• Dynamic normalization layer selection by @glenn-jocher in Dynamic normalization layer selection #7392
• Add version warning for wandb by @AyushExel in Add version warning for wandb #7385
• Remove OpenVINO ONNX opset<=12 check by @glenn-jocher in Remove OpenVINO ONNX opset<=12 check #7398
• Fix EdgeTPU output directory by @glenn-jocher in Fix EdgeTPU output directory #7399
• Update git_describe() by @glenn-jocher in Update git_describe() #7402
• Remove tensorrt pip install check by @glenn-jocher in Remove tensorrt pip install check #7439
• Disable pbar for DDP ranks > 0 by @glenn-jocher in Disable pbar for DDP ranks > 0 #7440
• Add --half support for FP16 CoreML exports with by @Cedric-Perauer in Add --half support for FP16 CoreML exports with #7446
• Bump cirrus-actions/rebase from 1.5 to 1.6 by @dependabot in Bump cirrus-actions/rebase from 1.5 to 1.6 #7462
• Update val.py by @HERIUN in Update val.py #7478
• Improved non-latin Annotator() plotting by @glenn-jocher in Improved non-latin Annotator() plotting #7488
• check_fonts() download to CONFIG_DIR fix by @glenn-jocher in check_fonts() download to CONFIG_DIR fix #7489
• Fix val.py Ensemble() by @glenn-jocher in Fix val.py Ensemble() #7490
• Added YOLOv5_AUTOINSTALL environment variable by @jkocherhans in Added YOLOv5_AUTOINSTALL environment variable #7505
• Refactor Dockerfiles to utils/docker by @glenn-jocher in Refactor Dockerfiles to utils/docker #7510
• Add yesqa to precommit checks by @Zengyf-CVer in Add yesqa to precommit checks #7511
• Fix val plots=plots by @glenn-jocher in Fix val plots=plots #7524
• Reduce val device transfers by @glenn-jocher in Reduce val device transfers #7525
• Add Docker --file argument to build by @glenn-jocher in Add Docker --file argument to build #7527
• Empty val batch CUDA device fix by @glenn-jocher in Empty val batch CUDA device fix #7539
• Autoinstall TensorRT if missing by @glenn-jocher in Autoinstall TensorRT if missing #7537
• Add mdformat to precommit checks and update other version by @Zengyf-CVer in Add mdformat to precommit checks and update other version #7529
• Update precommit monthly python 3.7+ by @glenn-jocher in Update precommit monthly python 3.7+ #7542
• Update downloads.py current release by @Powercube7 in Update downloads.py current release #7541
• Update check_requirements() with cmds=() argument by @glenn-jocher in Update check_requirements() with cmds=() argument #7543
• Add --noplots flag to suppress figures and images logging by @AyushExel in Add --noplots flag to suppress figures and images logging #7534
• Improve availability of YOLOv5 in Russia by @glenn-jocher in Improve availability of YOLOv5 in Russia #7545
• Update restapi.py by @glenn-jocher in Update restapi.py #7309
• Restrict TRT autoinstall to Linux-only by @glenn-jocher in Restrict TRT autoinstall to Linux-only #7549
• PyTorch Hub _verbose=False fix2 by @glenn-jocher in PyTorch Hub _verbose=False fix2 #7550
• Enable results.print() when _verbose=False by @glenn-jocher in Enable results.print() when _verbose=False #7558
• Add print(results) override for PyTorch Hub results by @glenn-jocher in Add print(results) override for PyTorch Hub results #7559
• TensorRT PyTorch Hub inference fix by @glenn-jocher in TensorRT PyTorch Hub inference fix #7560
• Add PyTorch-only benchmark arg by @glenn-jocher in Add PyTorch-only benchmark arg #7564
• Fix from yolov5 import utils statement by @Jack24658735 in Fix from yolov5 import utils statement #7578
• Remove tqdm.auto by @glenn-jocher in Remove tqdm.auto #7599
• Add --half support for OpenVINO exports by @djmmoss in Add --half support for OpenVINO exports #7615
• increment_path() robustness improvements by @glenn-jocher in increment_path() robustness improvements #7628
• Reduce opencv-python>=4.1.1 for Jetson Nano by @glenn-jocher in Reduce opencv-python>=4.1.1 for Jetson Nano #7645
• Bump scipy>=1.5 to meet numpy constraints by @glenn-jocher in Bump scipy>=1.5 to meet numpy constraints #7646
• Update plot.py by @sylvanding in Update plot.py #7654
• Bump github/codeql-action from 1 to 2 by @dependabot in Bump github/codeql-action from 1 to 2 #7665
• Disallow --dynamic when --half is passed by @dmatos2012 in Disallow --dynamic when --half is passed #7669
• Update Dockerfile-cpu to force python3.9 by @glenn-jocher in Update Dockerfile-cpu to force python3.9 #7675
• Pin Docker-cpu FROM ubuntu:20.04 by @glenn-jocher in Pin Docker-cpu FROM ubuntu:20.04 #7677
• FROM nvcr.io/nvidia/pytorch:22.04-py3 by @glenn-jocher in FROM nvcr.io/nvidia/pytorch:22.04-py3 #7680
• Update README.md by @glenn-jocher in Update README.md #7686
• Update tutorial.ipynb by @glenn-jocher in Update tutorial.ipynb #7715
• Update on_train_end callback by @glenn-jocher in Update on_train_end callback #7716
• Report fused model summary by default by @glenn-jocher in Report fused model summary by default #7722
• Ability to download older assets by @CristiFati in Ability to download older assets #7767
• Pin downloads to release version by @glenn-jocher in Pin downloads to release version #7790
• Implement DDP static_graph=True by @glenn-jocher in Implement DDP static_graph=True #6940
• Add smoothing to curve plots and max F1 index by @glenn-jocher in Add smoothing to curve plots and max F1 index #7798
• Add Dockerfile-M1 by @glenn-jocher in Add Dockerfile-M1 #7720
• Add random interpolation method augmentation by @developer0hye in Add random interpolation method augmentation #6826
• Rename utils/datasets.py -> utils/dataloaders.py by @glenn-jocher in Rename utils/datasets.py -> utils/dataloaders.py #7799
• Add @threaded decorator by @glenn-jocher in Add @threaded decorator #7813
• Add Python version output by @glenn-jocher in Add Python version output #7814
• Refactor collections and fstrings by @glenn-jocher in Refactor collections and fstrings #7821
• Refactor modules by @glenn-jocher in Refactor modules #7823
• YOLOv5 OpenVINO PyTorch Hub inference fix by @glenn-jocher in YOLOv5 OpenVINO PyTorch Hub inference fix #7826
• New TensorFlow TFDWConv() module by @glenn-jocher in New TensorFlow TFDWConv() module #7824
• Bump cirrus-actions/rebase from 1.6 to 1.7 by @dependabot in Bump cirrus-actions/rebase from 1.6 to 1.7 #7831
• New TensorFlow TFCrossConv() module by @glenn-jocher in New TensorFlow TFCrossConv() module #7827
• Fix TFDWConv() c1 == c2 check by @glenn-jocher in Fix TFDWConv() c1 == c2 check #7842
• Replace openvino-dev with OpenVINO Runtime inference by @GabrielDornelles in Replace openvino-dev with OpenVINO Runtime inference #7843
• TFDWConv() depthwise_initializer fix by @glenn-jocher in TFDWConv() depthwise_initializer fix #7845
• Update Dockerfile --no-install-recommends by @glenn-jocher in Update Dockerfile --no-install-recommends #7846
• Update CI CPU badge by @glenn-jocher in Update CI CPU badge #7855
• Create docker.yml by @glenn-jocher in Create docker.yml #7856
• Update Dockerfile-cpu install libpython3.8-dev by @glenn-jocher in Update Dockerfile-cpu install libpython3.8-dev #7857
• Add TFDWConv() depth_multiplier arg by @glenn-jocher in Add TFDWConv() depth_multiplier arg #7858
• Update Dockerfile-arm64 by @glenn-jocher in Update Dockerfile-arm64 #7860
• Invert Docker Image publishing order by @glenn-jocher in Invert Docker Image publishing order #7877
• Removed shell=True from subprocess commands that require user inputs by @JWLee89 in Removed shell=True from subprocess commands that require user inputs #7875
• Added Windows cmd to count GPU devices by @xylieong in Added Windows cmd to count GPU devices #7891
• Bug fix mAP0.5-0.95 by @lebedevdes in Bug fix mAP0.5-0.95 #6787
• cv2.resize interpolation fix by @glenn-jocher in cv2.resize interpolation fix #7903
• Add DWConvTranspose2d() module by @glenn-jocher in Add DWConvTranspose2d() module #7881
• Add check_yaml() to benchmarks.py by @glenn-jocher in Add check_yaml() to benchmarks.py #7916
• Add --keras argument for TF exports by @glenn-jocher in Add --keras argument for TF exports #7921
• Add PyTorch AMP check by @glenn-jocher in Add PyTorch AMP check #7917
• Code refactor by @glenn-jocher in Code refactor #7923
• AMP check image download backup by @glenn-jocher in AMP check image download backup #7936
• Fix AMP check tolerance by @glenn-jocher in Fix AMP check tolerance #7937
• Windows check_file() fix by @glenn-jocher in Windows check_file() fix #7938
• Add OpenVINO metadata to export by @xylieong in Add OpenVINO metadata to export #7947
• OpenVINO metadata fix by @glenn-jocher in OpenVINO metadata fix #7952
• OpenVINO metadata fix2 by @xylieong in OpenVINO metadata fix2 #7954
• YOLOv5 Apple Metal Performance Shader (MPS) support by @glenn-jocher in YOLOv5 Apple Metal Performance Shader (MPS) support #7878
• Increase NMS time limit to 50 ms/img by @glenn-jocher in Increase NMS time limit to 50 ms/img #7956
• Increase NMS fixed time limit 300ms + 30ms/img by @glenn-jocher in Increase NMS fixed time limit 300ms + 30ms/img #7957
• AMP check improvements backup YOLOv5n pretrained by @glenn-jocher in AMP check improvements backup YOLOv5n pretrained #7959
• Update greetings.yml by @glenn-jocher in Update greetings.yml #7965
• Update requirements.txt with Usage example by @glenn-jocher in Update requirements.txt with Usage example #7966
• Remove formats variable to avoid pd conflict by @glenn-jocher in Remove formats variable to avoid pd conflict #7993
• check_requirements() Windows fix by @glenn-jocher in check_requirements() Windows fix #7997
• Add psutil and ipython to requirements.txt by @glenn-jocher in Add psutil and ipython to requirements.txt #7998
• cURL download double quotes by @glenn-jocher in cURL download double quotes #7999
• Update and add benchmarks to ci-testing.yml by @glenn-jocher in Update and add benchmarks to ci-testing.yml #7996
• Add warning emoji ⚠️ on --conf > 0.001 by @glenn-jocher in Add warning emoji ⚠️ on --conf > 0.001 #8005
• Update CI benchmarks to ONNX autoinstall by @glenn-jocher in Update CI benchmarks to ONNX autoinstall #8008
• Pin CI protobuf==3.20.1 by @glenn-jocher in Pin CI protobuf==3.20.1 #8015
• check_requirements() "" Windows double quote by @glenn-jocher in check_requirements() "" Windows double quote #8016
• Add requirements.txt protobuf<=3.20.1 by @glenn-jocher in Add requirements.txt protobuf<=3.20.1 #8017
• Add best.pt PyTorch Hub inference to CI by @glenn-jocher in Add best.pt PyTorch Hub inference to CI #8024
• Add Dockerfile descriptions by @glenn-jocher in Add Dockerfile descriptions #8031
• Simplify and fix --save-period epoch 0 by @glenn-jocher in Simplify and fix --save-period epoch 0 #8042
• .detach() on bias init by @tcmyxc in .detach() on bias init #8044
• attempt_load() deserialize fix by @glenn-jocher in attempt_load() deserialize fix #8051
• Fix FP32 TensorRT model export by @SahilChachra in Fix FP32 TensorRT model export #8046
• Make Docker actions conditional on ultralytics/yolov5 repo by @glenn-jocher in Make Docker actions conditional on ultralytics/yolov5 repo #8060
• Update stale.yml to 300 ops per run by @glenn-jocher in Update stale.yml to 300 ops per run #8061
• Fix torch long to float tensor on HUB macOS by @glenn-jocher in Fix torch long to float tensor on HUB macOS #8067
• Improved dataset error introspection by @glenn-jocher in Improved dataset error introspection #8091
• experimental.py Apple MPS device fix by @glenn-jocher in experimental.py Apple MPS device fix #8121
• [pre-commit.ci] pre-commit suggestions by @pre-commit-ci in [pre-commit.ci] pre-commit suggestions #8119
• fix(utils): missing edge when resample segments by @HRan2004 in fix(utils): missing edge when resample segments #8092
• AutoBatch checks against failed solutions by @glenn-jocher in AutoBatch checks against failed solutions #8159
• FROM nvcr.io/nvidia/pytorch:22.05-py3 by @glenn-jocher in FROM nvcr.io/nvidia/pytorch:22.05-py3 #8162
• Docker GPU pip install opencv-python<4.6.0.66 by @glenn-jocher in Docker GPU pip install opencv-python<4.6.0.66 #8164
• Improved dataset_stats() YAML checks by @glenn-jocher in Improved dataset_stats() YAML checks #8125
• Bump actions/setup-python from 3 to 4 by @dependabot in Bump actions/setup-python from 3 to 4 #8186
• Reject webcam inference on Colab/Kaggle by @glenn-jocher in Reject webcam inference on Colab/Kaggle #8209
• Remove streaming source sleep period by @glenn-jocher in Remove streaming source sleep period #8243
• Prefer MPS over CPU if available by @glenn-jocher in Prefer MPS over CPU if available #8210
• HUB dataset_stats() error reporting by @glenn-jocher in HUB dataset_stats() error reporting #8192
• process_batch() as numpy arrays by @glenn-jocher in process_batch() as numpy arrays #8254
• Created using Colaboratory by @glenn-jocher in Created using Colaboratory #8255
• logging.ERROR on RANK not in (0, 1) by @glenn-jocher in logging.ERROR on RANK not in (0, 1) #8284
• device_name="MYRIAD" for Intel NCS2 comment by @glenn-jocher in device_name="MYRIAD" for Intel NCS2 comment #8327
• Allow preview resize in detect.py by @RyanHir in Allow preview resize in detect.py #8318
• Create README_cn.md by @glenn-jocher in Create README_cn.md #8344
• Allow detect.py to use video size for initial window size by @NicholasZolton in Allow detect.py to use video size for initial window size #8330
• Revamp Chinese docs by @zhiqwang in Revamp Chinese docs #8350
• Fix bias warmup LR init by @glenn-jocher in Fix bias warmup LR init #8356
• Add File Size (MB) column to benchmarks by @glenn-jocher in Add File Size (MB) column to benchmarks #8359
• Update protobuf requirement from <=3.20.1 to <4.21.3 by @dependabot in Update protobuf requirement from <=3.20.1 to <4.21.3 #8346
• Fix ONNX --dynamic export on GPU by @glenn-jocher in Fix ONNX --dynamic export on GPU #8378
• Update tutorial.ipynb by @glenn-jocher in Update tutorial.ipynb #8380
• Implementation of Early Stopping for DDP training by @giacomoguiduzzi in Implementation of Early Stopping for DDP training #8345
• Improve --local_rank arg comment by @pourmand1376 in Improve --local_rank arg comment #8409
• Update cache comments by @glenn-jocher in Update cache comments #8414
• TRT --half fix autocast images to FP16 by @glenn-jocher in TRT --half fix autocast images to FP16 #8435
• Expose OpenVINO batch_size similarly to TensorRT by @democat3457 in Expose OpenVINO batch_size similarly to TensorRT #8437
• Do not prefer Apple MPS by @glenn-jocher in Do not prefer Apple MPS #8446
• Update stale.yml by @glenn-jocher in Update stale.yml #8465
• [pre-commit.ci] pre-commit suggestions by @pre-commit-ci in [pre-commit.ci] pre-commit suggestions #8470
• Exclude torch==1.12.0, torchvision==0.13.0 (Fix YOLOv5 issues with torch==1.12 on Multi-GPU systems #8395) by @mjun0812 in Exclude torch==1.12.0, torchvision==0.13.0 (Fix #8395) #8497
• Update tutorial.ipynb by @glenn-jocher in Update tutorial.ipynb #8507
• Training reproducibility improvements by @AyushExel in Training reproducibility improvements #8213
• Revert "Expose OpenVINO batch_size similarly to TensorRT" by @glenn-jocher in Revert "Expose OpenVINO batch_size similarly to TensorRT" #8510
• Avoid FP64 ops for MPS support in train.py by @glenn-jocher in Avoid FP64 ops for MPS support in train.py #8511
• Fix ap calculation bug Val.py: No prediction on image follows in wrong AP calculation #8464 by @UnglvKitDe in Fix ap calculation bug #8464 #8484
• Add --hard-fail argument to benchmarks for CI errors by @glenn-jocher in Add --hard-fail argument to benchmarks for CI errors #8513
• Simplify benchmarks.py assertions by @glenn-jocher in Simplify benchmarks.py assertions #8515
• Properly expose batch_size from OpenVINO similarly to TensorRT by @democat3457 in Properly expose batch_size from OpenVINO similarly to TensorRT #8514
• Add --half arguments to export.py Usage examples by @glenn-jocher in Add --half arguments to export.py Usage examples #8516
• XML export --half fix by @glenn-jocher in XML export --half fix #8522
• Fix LoadImages() with dataset YAML lists by @democat3457 in Fix LoadImages() with dataset YAML lists #8517
• val.py assert ncm == nc fix by @glenn-jocher in val.py assert ncm == nc fix #8545
• CIoU protected divides by @glenn-jocher in CIoU protected divides #8546
• Update metrics.py with IoU protected divides by @glenn-jocher in Update metrics.py with IoU protected divides #8550
• Add TensorRT dependencies by @Zengyf-CVer in Add TensorRT dependencies #8553
• Add thop>=0.1.0 by @glenn-jocher in Add thop>=0.1.0 #8558
• Raise error on suffix-less model path by @democat3457 in Raise error on suffix-less model path #8561
• Assert --optimize not used with cuda device by @democat3457 in Assert --optimize not used with cuda device #8569
• Update requirements.txt comment spacing by @glenn-jocher in Update requirements.txt comment spacing #8562
• Explicitly set weight_decay value by @developer0hye in Explicitly set weight_decay value #8592
• Update scipy>=1.7.3 by @glenn-jocher in Update scipy>=1.7.3 #8595
• Update tqdm>=4.64.0 and thop>=0.1.1 by @glenn-jocher in Update tqdm>=4.64.0 and thop>=0.1.1 #8596
• Add pip install wheel to avoid legacy setup.py install by @glenn-jocher in Add pip install wheel to avoid legacy setup.py install #8597
• Link fuse() to AutoShape() for Hub models by @glenn-jocher in Link fuse() to AutoShape() for Hub models #8599
• FROM nvcr.io/nvidia/pytorch:22.06-py3 by @glenn-jocher in FROM nvcr.io/nvidia/pytorch:22.06-py3 #8600
• Refactor optimizer initialization by @glenn-jocher in Refactor optimizer initialization #8607
• assert torch!=1.12.0 for DDP training by @glenn-jocher in assert torch!=1.12.0 for DDP training #8621
• Fix OMP_NUM_THREADS=1 for macOS by @glenn-jocher in Fix OMP_NUM_THREADS=1 for macOS #8624
• Upgrade onnxsim to v0.4.1 by @daquexian in Upgrade onnxsim to v0.4.1 #8632
• Check TensorBoard logger before adding graph by @JarnoRFB in Check TensorBoard logger before adding graph #8664
• Use contextlib's suppress method to silence an error by @glenn-jocher in Use contextlib's suppress method to silence an error #8668
• Update hubconf.py to reset LOGGER.level after load by @glenn-jocher in Update hubconf.py to reset LOGGER.level after load #8674
• Update warning emojis by @glenn-jocher in Update warning emojis #8678
• Update hubconf.py to reset logging level to INFO by @glenn-jocher in Update hubconf.py to reset logging level to INFO #8680
• Add generator and worker seed by @UnglvKitDe in Add generator and worker seed #8602
• Set torch.cuda.manual_seed_all() for DDP by @glenn-jocher in Set torch.cuda.manual_seed_all() for DDP #8688
• Move .dockerignore to root by @glenn-jocher in Move .dockerignore to root #8690
• Faster crop saving by @glenn-jocher in Faster crop saving #8696
• Remove else: from load_image() by @glenn-jocher in Remove else: from load_image() #8692
• Avoid cv2 window init code on Windows by @glenn-jocher in Avoid cv2 window init code on Windows #8712
• Update dataloaders.py by @glenn-jocher in Update dataloaders.py #8714
• New HUBDatasetStats() class by @glenn-jocher in New HUBDatasetStats() class #8716
• Fix BGR->RGB Bug in albumentations Albumentations-Pipeline is applied to BGR not to RGB #8641 by @UnglvKitDe in Fix BGR->RGB Bug in albumentations #8641 #8695
• coremltools>=5.2 for CoreML export by @glenn-jocher in coremltools>=5.2 for CoreML export #8725
• Revert "Fix BGR->RGB Bug in albumentations Albumentations-Pipeline is applied to BGR not to RGB #8641" by @glenn-jocher in Revert "Fix BGR->RGB Bug in albumentations #8641" #8727
• fix: broken is_docker check by @maxstrobel in fix: broken is_docker check #8711
• Revert protobuf<=3.20.1 by @glenn-jocher in Revert protobuf<=3.20.1 #8742
• Dynamic batch size support for TensorRT by @democat3457 in Dynamic batch size support for TensorRT #8526
• Fix confusion matrix update when no predictions are made by @jbutle55 in Fix confusion matrix update when no predictions are made #8748
• Add val.py no label warning by @glenn-jocher in Add val.py no label warning #8782
• Fix detect.py --update list bug by @glenn-jocher in Fix detect.py --update list bug #8783
• ci-testing.yml Windows-friendly ENV variables by @glenn-jocher in ci-testing.yml Windows-friendly ENV variables #8794
• Add hubconf.py argparser by @glenn-jocher in Add hubconf.py argparser #8799
• Print hubconf.py args by @glenn-jocher in Print hubconf.py args #8800
• Update Colab Notebook CI by @glenn-jocher in Update Colab Notebook CI #8798
• Deprecate torch 1.6.0 compat _non_persistent_buffers_set by @glenn-jocher in Deprecate torch 1.6.0 compat _non_persistent_buffers_set #8797
• Detect.inplace=False for multithread-safe inference by @glenn-jocher in Detect.inplace=False for multithread-safe inference #8801
• Update train.py for val.run(half=amp) by @glenn-jocher in Update train.py for val.run(half=amp) #8804
• Fix val.py 'no labels found bug' by @glenn-jocher in Fix val.py 'no labels found bug' #8806
• Update requirements.txt with tf-cpu and tf-aarch64 by @glenn-jocher in Update requirements.txt with tf-cpu and tf-aarch64 #8807
• FROM nvcr.io/nvidia/pytorch:22.07-py3 by @glenn-jocher in FROM nvcr.io/nvidia/pytorch:22.07-py3 #8808
• Update ci-testing.yml streamlined tests by @glenn-jocher in Update ci-testing.yml streamlined tests #8809
• Check git status on upstream ultralytics or origin dynamically by @pourmand1376 in Check git status on upstream ultralytics or origin dynamically #8694
• Fix Colab-update pre-commit EOF bug by @glenn-jocher in Fix Colab-update pre-commit EOF bug #8810
• Update .pre-commit-config.yaml by @glenn-jocher in Update .pre-commit-config.yaml #8811
• Update .pre-commit-config.yaml by @glenn-jocher in Update .pre-commit-config.yaml #8812
• Remove assert not is_docker() from GitHub checks by @glenn-jocher in Remove assert not is_docker() from GitHub checks #8813
• Add .git to .dockerignore by @glenn-jocher in Add .git to .dockerignore #8815
• Add tensor hooks and 10.0 gradient clipping by @UnglvKitDe in Add tensor hooks and 10.0 gradient clipping #8598
• Update README.md with contributors.png by @glenn-jocher in Update README.md with contributors.png #8820
• Remove hook torch.nan_to_num(x) by @glenn-jocher in Remove hook torch.nan_to_num(x) #8826
• RUN git clone instead of COPY to /usr/src/app by @glenn-jocher in RUN git clone instead of COPY to /usr/src/app #8827
• [pre-commit.ci] pre-commit suggestions by @pre-commit-ci in [pre-commit.ci] pre-commit suggestions #8828
• Fix missing attr model.model when loading custom yolov model by @jacksonargo in Fix missing attr model.model when loading custom yolov model #8830
• New smart_resume() by @glenn-jocher in New smart_resume() #8838
• Self-contained checkpoint --resume by @glenn-jocher in Self-contained checkpoint --resume #8839
• Add check_file(data) i.e. --data coco128.yaml by @glenn-jocher in Add check_file(data) i.e. --data coco128.yaml #8851
• export.py replace check_file -> check_yaml by @glenn-jocher in export.py replace check_file -> check_yaml #8852
• Update dataloaders.py remove float64 shapes by @glenn-jocher in Update dataloaders.py remove float64 shapes #8865
• Fix TensorRT --dynamic excess outputs bug by @glenn-jocher in Fix TensorRT --dynamic excess outputs bug #8869
• Single-line TRT dynamic assertion by @glenn-jocher in Single-line TRT dynamic assertion #8871
• HUBDatasetStats() preview images to 50 quality by @glenn-jocher in HUBDatasetStats() preview images to 50 quality #8880
• --resume training from URL weights by @glenn-jocher in --resume training from URL weights #8882
• --resume training from URL weights fix by @glenn-jocher in --resume training from URL weights fix #8884
• Update CI to default Python 3.10 by @glenn-jocher in Update CI to default Python 3.10 #8883
• ClearML experiment tracking integration by @thepycoder in ClearML experiment tracking integration #8620
• smart_optimizer() improved reporting by @glenn-jocher in smart_optimizer() improved reporting #8887
• emoji-safe default logging by @glenn-jocher in emoji-safe default logging #8888
• Fix/improve social media bar by @kalenmike in Fix/improve social media bar #8893
• Update README.md Environments icons by @glenn-jocher in Update README.md Environments icons #8895
• Fix anchor incorrectly initialized by @YellowAndGreen in Fix anchor incorrectly initialized #8891
• Edge TPU add --search_delegate fix by @glenn-jocher in Edge TPU add --search_delegate fix #8902
• Capitalize YouTube by @glenn-jocher in Capitalize YouTube #8903
• About "os.R_OK" in general.py by @19190205wzy in About "os.R_OK" in general.py #8909
• Scope torchvision.utils.draw_bounding_boxes import by @glenn-jocher in Scope torchvision.utils.draw_bounding_boxes import #8915
• Add torch==1.7.0 to CI tests by @glenn-jocher in Add torch==1.7.0 to CI tests #8916
• Direct download table links by @glenn-jocher in Direct download table links #8917
• Fix --source 'path/*.jpg' glob example by @glenn-jocher in Fix --source 'path/*.jpg' glob example #8927
• Deci.ai optimization and deployment Integration by @glenn-jocher in Deci.ai optimization and deployment Integration #8928
• Allow train to use zipped content by @kalenmike in Allow train to use zipped content #8940
• Remove DDP destroy_process_group() on train end by @glenn-jocher in Remove DDP destroy_process_group() on train end #8935
• GFLOPs computation fix for classification models by @glenn-jocher in GFLOPs computation fix for classification models #8954
• New smart_inference_mode() conditional decorator by @glenn-jocher in New smart_inference_mode() conditional decorator #8957
• Fix windows LOGGER with emojis output by @glenn-jocher in Fix windows LOGGER with emojis output #8958
• Add bash script arguments by @glenn-jocher in Add bash script arguments #8962
• Add .tar support for datasets by @glenn-jocher in Add .tar support for datasets #8963
• Allow multiple-model serving from Flask REST API by @mpromonet in Allow multiple-model serving from Flask REST API #8973
• Colab and Kaggle loggers removeHandler by @glenn-jocher in Colab and Kaggle loggers removeHandler #8985
• Fix bug with resume by @savchenkoyana in Fix bug with resume #8912

😃 New Contributors (41)

• @lcombaldieu made their first contribution in Fix export for 1-channel images #6780
• @vnekat made their first contribution in Update yolov5s.yaml #6865
• @DavidBaldsiefen made their first contribution in Default FP16 TensorRT export #6798
• @paradigmn made their first contribution in EdgeTPU optimizations #6808
• @maxstrobel made their first contribution in fix: add default PIL font as fallback  #7010
• @LaserBorg made their first contribution in Use PIL to eliminate chroma subsampling in crops #7008
• @yeshanliu made their first contribution in np.fromfile() Chinese image paths fix #6979
• @CCRcmcpe made their first contribution in Fix cv2.imwrite on non-ASCII paths #7139
• @leeflix made their first contribution in Enable TensorFlow ops for --nms and --agnostic-nms #7281
• @n1mmy made their first contribution in Copy wandb param dict before training to avoid overwrites. #7317
• @rglkt made their first contribution in DetectMultiBackend() default stride=32 #7342
• @vardanagarwal made their first contribution in Add support for different normalization layers #7377
• @Cedric-Perauer made their first contribution in Add --half support for FP16 CoreML exports with #7446
• @HERIUN made their first contribution in Update val.py #7478
• @jkocherhans made their first contribution in Added YOLOv5_AUTOINSTALL environment variable #7505
• @Powercube7 made their first contribution in Update downloads.py current release #7541
• @Jack24658735 made their first contribution in Fix from yolov5 import utils statement #7578
• @djmmoss made their first contribution in Add --half support for OpenVINO exports #7615
• @sylvanding made their first contribution in Update plot.py #7654
• @dmatos2012 made their first contribution in Disallow --dynamic when --half is passed #7669
• @GabrielDornelles made their first contribution in Replace openvino-dev with OpenVINO Runtime inference #7843
• @JWLee89 made their first contribution in Removed shell=True from subprocess commands that require user inputs #7875
• @xylieong made their first contribution in Added Windows cmd to count GPU devices #7891
• @lebedevdes made their first contribution in Bug fix mAP0.5-0.95 #6787
• @tcmyxc made their first contribution in .detach() on bias init #8044
• @SahilChachra made their first contribution in Fix FP32 TensorRT model export #8046
• @HRan2004 made their first contribution in fix(utils): missing edge when resample segments #8092
• @RyanHir made their first contribution in Allow preview resize in detect.py #8318
• @NicholasZolton made their first contribution in Allow detect.py to use video size for initial window size #8330
• @giacomoguiduzzi made their first contribution in Implementation of Early Stopping for DDP training #8345
• @pourmand1376 made their first contribution in Improve --local_rank arg comment #8409
• @democat3457 made their first contribution in Expose OpenVINO batch_size similarly to TensorRT #8437
• @mjun0812 made their first contribution in Exclude torch==1.12.0, torchvision==0.13.0 (Fix #8395) #8497
• @UnglvKitDe made their first contribution in Fix ap calculation bug #8464 #8484
• @daquexian made their first contribution in Upgrade onnxsim to v0.4.1 #8632
• @JarnoRFB made their first contribution in Check TensorBoard logger before adding graph #8664
• @jbutle55 made their first contribution in Fix confusion matrix update when no predictions are made #8748
• @jacksonargo made their first contribution in Fix missing attr model.model when loading custom yolov model #8830
• @thepycoder made their first contribution in ClearML experiment tracking integration #8620
• @YellowAndGreen made their first contribution in Fix anchor incorrectly initialized #8891
• @19190205wzy made their first contribution in About "os.R_OK" in general.py #8909
• @mpromonet made their first contribution in Allow multiple-model serving from Flask REST API #8973
• @savchenkoyana made their first contribution in Fix bug with resume #8912

[sotaDetector]

will you add keypoint task in this project later?

[AyushExel]

There's a lot of things coming. Stay tuned :)

[daikankan]

[daikankan]

@glenn-jocher referring to detection training

[glenn-jocher]

@daikankan detection training is unaffected by v6.2

[daikankan]

@glenn-jocher After I disabled [self.tb, self.clearml, self.wandb] in on_fit_epoch_end(), and on_model_save(), It's still stuck in early epochs.

[AyushExel]

@daikankan hey can you please share the original training command that got stuck? I'd like to test it to make sure we don't have a bug specific to large datasets. Thanks!

[daikankan]

@AyushExel nohup python -m torch.distributed.launch --nproc_per_node 4 train.py --workers 8 --cfg models/mymodel.yaml --batch-size 2048 --epochs 300 --noautoanchor --device 0,1,2,3 > ../log.out 2>&1 &

[rkechols]

These release notes refer to 6.3:

Our next release, v6.3 is scheduled for September

It is now late November. Is there any chance of a new tag/release coming soon?
My organization would like to make use of some of the recent changes (newer than 6.2), but don't want our code to pull the always-moving "latest from master", for stability reasons. (Note that torch.hub.load only allows pinning by branch or tag, not by commit SHA)

@glenn-jocher

[glenn-jocher]

Tomorrow. #10245