v6.1 - TensorRT, TensorFlow Edge TPU and OpenVINO Export and Inference #6740

glenn-jocher · 2022-02-22T11:35:58Z

glenn-jocher
Feb 22, 2022
Maintainer

This release incorporates new features and bug fixes (271 PRs from 48 contributors) since our last release in October 2021. It adds TensorRT, Edge TPU and OpenVINO support, and provides retrained models at --batch-size 128 with new default one-cycle linear LR scheduler. YOLOv5 now officially supports 11 different formats, not just for export but for inference (both detect.py and PyTorch Hub), and validation to profile mAP and speed results after export.

Format	`export.py --include`	Model
PyTorch	-	`yolov5s.pt`
TorchScript	`torchscript`	`yolov5s.torchscript`
ONNX	`onnx`	`yolov5s.onnx`
OpenVINO	`openvino`	`yolov5s_openvino_model/`
TensorRT	`engine`	`yolov5s.engine`
CoreML	`coreml`	`yolov5s.mlmodel`
TensorFlow SavedModel	`saved_model`	`yolov5s_saved_model/`
TensorFlow GraphDef	`pb`	`yolov5s.pb`
TensorFlow Lite	`tflite`	`yolov5s.tflite`
TensorFlow Edge TPU	`edgetpu`	`yolov5s_edgetpu.tflite`
TensorFlow.js	`tfjs`	`yolov5s_web_model/`

Usage examples (ONNX shown):

Export:          python export.py --weights yolov5s.pt --include onnx
Detect:          python detect.py --weights yolov5s.onnx
PyTorch Hub:     model = torch.hub.load('ultralytics/yolov5', 'custom', 'yolov5s.onnx')
Validate:        python val.py --weights yolov5s.onnx
Visualize:       https://netron.app

Important Updates

TensorRT support: TensorFlow, Keras, TFLite, TF.js model export now fully integrated using python export.py --include saved_model pb tflite tfjs (Export, detect and validation with TensorRT engine file #5699 by @imyhxy)
Tensorflow Edge TPU support ⭐ NEW: New smaller YOLOv5n (1.9M params) model below YOLOv5s (7.5M params), exports to 2.1 MB INT8 size, ideal for ultralight mobile solutions. (Add EdgeTPU support #3630 by @zldrobit)
OpenVINO support: YOLOv5 ONNX models are now compatible with both OpenCV DNN and ONNX Runtime (OpenVINO Export #6057 by @glenn-jocher).
Export Benchmarks: Benchmark (mAP and speed) all YOLOv5 export formats with python utils/benchmarks.py --weights yolov5s.pt. Currently operates on CPU, future updates will implement GPU support. (YOLOv5 CPU Export Benchmarks #6613 by @glenn-jocher).
Architecture: no changes
Hyperparameters: minor change
- hyp-scratch-large.yaml lrf reduced from 0.2 to 0.1 (Update hyp.scratch-high.yaml lrf: 0.1 #6525 by @glenn-jocher).
Training: Default Learning Rate (LR) scheduler updated
- One-cycle with cosine replace with one-cycle linear for improved results (Switch default LR scheduler from cos to linear #6729 by @glenn-jocher).

New Results

All model trainings logged to https://wandb.ai/glenn-jocher/YOLOv5_v61_official

YOLOv5-P5 640 Figure (click to expand)

Figure Notes (click to expand)

COCO AP val denotes mAP@0.5:0.95 metric measured on the 5000-image COCO val2017 dataset over various inference sizes from 256 to 1536.
GPU Speed measures average inference time per image on COCO val2017 dataset using a AWS p3.2xlarge V100 instance at batch-size 32.
EfficientDet data from google/automl at batch size 8.
Reproduce by python val.py --task study --data coco.yaml --iou 0.7 --weights yolov5n6.pt yolov5s6.pt yolov5m6.pt yolov5l6.pt yolov5x6.pt

Example YOLOv5l before and after metrics:

YOLOv5l ^Large	size ^(pixels)	mAP^val 0.5:0.95	mAP^val 0.5	Speed ^{CPU b1 (ms)}	Speed ^{V100 b1 (ms)}	Speed ^{V100 b32 (ms)}	params ^(M)	FLOPs ^{@640 (B)}
v5.0	640	48.2	66.9	457.9	11.6	2.8	47.0	115.4
v6.0 (previous)	640	48.8	67.2	424.5	10.9	2.7	46.5	109.1
v6.1 (this release)	640	49.0	67.3	424.5	10.9	2.7	46.5	109.1

Pretrained Checkpoints

Model	size ^(pixels)	mAP^val 0.5:0.95	mAP^val 0.5	Speed ^{CPU b1 (ms)}	Speed ^{V100 b1 (ms)}	Speed ^{V100 b32 (ms)}	params ^(M)	FLOPs ^{@640 (B)}
YOLOv5n	640	28.0	45.7	45	6.3	0.6	1.9	4.5
YOLOv5s	640	37.4	56.8	98	6.4	0.9	7.2	16.5
YOLOv5m	640	45.4	64.1	224	8.2	1.7	21.2	49.0
YOLOv5l	640	49.0	67.3	430	10.1	2.7	46.5	109.1
YOLOv5x	640	50.7	68.9	766	12.1	4.8	86.7	205.7

YOLOv5n6	1280	36.0	54.4	153	8.1	2.1	3.2	4.6
YOLOv5s6	1280	44.8	63.7	385	8.2	3.6	16.8	12.6
YOLOv5m6	1280	51.3	69.3	887	11.1	6.8	35.7	50.0
YOLOv5l6	1280	53.7	71.3	1784	15.8	10.5	76.8	111.4
YOLOv5x6 + TTA	1280 1536	55.0 55.8	72.7 72.7	3136 -	26.2 -	19.4 -	140.7 -	209.8 -

Table Notes (click to expand)

All checkpoints are trained to 300 epochs with default settings. Nano and Small models use hyp.scratch-low.yaml hyps, all others use hyp.scratch-high.yaml.
mAP^val values are for single-model single-scale on COCO val2017 dataset.
Reproduce by python val.py --data coco.yaml --img 640 --conf 0.001 --iou 0.65
Speed averaged over COCO val images using a AWS p3.2xlarge instance. NMS times (~1 ms/img) not included.
Reproduce by python val.py --data coco.yaml --img 640 --task speed --batch 1
TTA Test Time Augmentation includes reflection and scale augmentations.
Reproduce by python val.py --data coco.yaml --img 1536 --iou 0.7 --augment

Changelog

Changes between previous release and this release: v6.0...v6.1
Changes since this release: v6.1...HEAD

New Features and Bug Fixes (271)

fix tf conversion in new v6 models by @YoniChechik in fix tf conversion in new v6 models #5153
Use YOLOv5n for CI testing by @glenn-jocher in Use YOLOv5n for CI testing #5154
Update stale.yml by @glenn-jocher in Update stale.yml #5156
Check 'onnxruntime-gpu' if torch.has_cuda by @glenn-jocher in Check 'onnxruntime-gpu' if torch.has_cuda #5087
Add class filtering to LoadImagesAndLabels() dataloader by @glenn-jocher in Add class filtering to LoadImagesAndLabels() dataloader #5172
W&B: fix dpp with wandb disabled by @AyushExel in W&B: fix dpp with wandb disabled #5163
Update autodownload fallbacks to v6.0 assets by @glenn-jocher in Update autodownload fallbacks to v6.0 assets #5177
W&B: DDP fix by @AyushExel in W&B: DDP fix #5176
Adjust legend labels for classes without instances by @NauchtanRobotics in Adjust legend labels for classes without instances #5174
Improved check_suffix() robustness to '' and "" by @glenn-jocher in Improved check_suffix() robustness to '' and "" #5192
Highlight contributors in README by @glenn-jocher in Highlight contributors in README #5173
Add hyp.scratch-med.yaml by @glenn-jocher in Add hyp.scratch-med.yaml #5196
Update Objects365.yaml to include the official validation set by @farleylai in Update Objects365.yaml to include the official validation set #5194
Autofix duplicate label handling by @glenn-jocher in Autofix duplicate label handling #5210
Update Objects365.yaml val count by @glenn-jocher in Update Objects365.yaml val count #5212
Update/inplace ops by @glenn-jocher in Update/inplace ops #5233
Add on_fit_epoch_end callback by @glenn-jocher in Add on_fit_epoch_end callback #5232
Update rebase.yml by @glenn-jocher in Update rebase.yml #5245
Add dependabot for GH actions by @zhiqwang in Add dependabot for GH actions #5250
Bump cirrus-actions/rebase from 1.4 to 1.5 by @dependabot in Bump cirrus-actions/rebase from 1.4 to 1.5 #5251
Bump actions/cache from 1 to 2.1.6 by @dependabot in Bump actions/cache from 1 to 2.1.6 #5252
Bump actions/stale from 3 to 4 by @dependabot in Bump actions/stale from 3 to 4 #5253
Update rebase.yml with workflows permissions by @glenn-jocher in Update rebase.yml with workflows permissions #5255
autosplit: take image files with uppercase extensions into account by @jdfr in autosplit: take image files with uppercase extensions into account #5269
take EXIF orientation tags into account when fixing corrupt images by @jdfr in take EXIF orientation tags into account when fixing corrupt images #5270
More informative EarlyStopping() message by @glenn-jocher in More informative EarlyStopping() message #5303
Uncomment OpenCV 4.5.4 requirement in detect.py by @SamFC10 in Uncomment OpenCV 4.5.4 requirement in detect.py #5305
Weights download script minor improvements by @CristiFati in Weights download script minor improvements #5213
Small fixes to docstrings by @zhiqwang in Small fixes to docstrings #5313
W&B: Media panel fix by @AyushExel in W&B: Media panel fix #5317
Add autobatch feature for best batch-size estimation by @glenn-jocher in Add autobatch feature for best batch-size estimation #5092
Update AutoShape.forward() model.classes example by @glenn-jocher in Update AutoShape.forward() model.classes example #5324
DDP nl fix by @glenn-jocher in DDP nl fix #5332
Add pre-commit CI action by @Borda in Add pre-commit CI action #4982
W&B: Fix sweep by @AyushExel in W&B: Fix sweep #5402
Update GitHub issues templates by @glenn-jocher in Update GitHub issues templates #5404
Fix MixConv2d() remove shortcut + apply depthwise by @glenn-jocher in Fix MixConv2d() remove shortcut + apply depthwise #5410
Meshgrid indexing='ij' for PyTorch 1.10 by @glenn-jocher in Meshgrid indexing='ij' for PyTorch 1.10 #5309
Update get_loggers() by @glenn-jocher in Update get_loggers() #4854
Update README.md by @glenn-jocher in Update README.md #5438
Fixed a small typo in CONTRIBUTING.md by @pranathlcp in Fixed a small typo in CONTRIBUTING.md #5445
Update check_git_status() to run under ROOT working directory by @MrinalJain17 in Update check_git_status() to run under ROOT working directory #5441
Fix tf.py LoadImages() dataloader return values by @glenn-jocher in Fix tf.py LoadImages() dataloader return values #5455
Remove check_requirements(('tensorflow>=2.4.1',)) by @glenn-jocher in Remove check_requirements(('tensorflow>=2.4.1',)) #5476
Improve GPU name by @glenn-jocher in Improve GPU name #5478
Update torch_utils.py import LOGGER by @glenn-jocher in Update torch_utils.py import LOGGER #5483
Add tf.py verification printout by @glenn-jocher in Add tf.py verification printout #5484
Keras CI fix by @glenn-jocher in Keras CI fix #5486
Delete code-format.yml by @glenn-jocher in Delete code-format.yml #5487
Fix float zeros format by @glenn-jocher in Fix float zeros format #5491
Handle edgetpu model inference by @Namburger in Handle edgetpu model inference #5372
precommit: isort by @Borda in precommit: isort #5493
Fix increment_path() with multiple-suffix filenames by @glenn-jocher in Fix increment_path() with multiple-suffix filenames #5518
Write date in checkpoint file by @developer0hye in Write date in checkpoint file #5514
Update plots.py feature_visualization path issues by @ys31jp in Update plots.py feature_visualization path issues #5519
Update cls bias init by @glenn-jocher in Update cls bias init #5520
Common is_coco logic betwen train.py and val.py by @glenn-jocher in Common is_coco logic betwen train.py and val.py #5521
Fix increment_path() explicit file vs dir handling by @glenn-jocher in Fix increment_path() explicit file vs dir handling #5523
Fix detect.py URL inference by @glenn-jocher in Fix detect.py URL inference #5525
Update check_file() avoid repeat URL downloads by @glenn-jocher in Update check_file() avoid repeat URL downloads #5526
Update export.py by @nanmi in Update export.py #5471
Update train.py by @wonbeomjang in Update train.py #5451
Suppress ONNX export trace warning by @deepsworld in Suppress ONNX export trace warning #5437
Update autobatch.py by @glenn-jocher in Update autobatch.py #5536
Update autobatch.py by @glenn-jocher in Update autobatch.py #5538
Update Issue Templates with 💡 ProTip! by @glenn-jocher in Update Issue Templates with 💡 ProTip! #5539
Update models/hub/*.yaml files for v6.0n release by @glenn-jocher in Update models/hub/*.yaml files for v6.0n release #5540
intersect_dicts() in hubconf.py fix by @glenn-jocher in intersect_dicts() in hubconf.py fix #5542
Fix for *.yaml emojis on load by @glenn-jocher in Fix for *.yaml emojis on load #5543
Fix save_one_box() by @glenn-jocher in Fix save_one_box() #5545
Inside Ultralytics video https://youtu.be/Zgi9g1ksQHc by @glenn-jocher in Inside Ultralytics video https://youtu.be/Zgi9g1ksQHc #5546
Add --conf-thres >> 0.001 warning by @glenn-jocher in Add --conf-thres >> 0.001 warning #5567
LOGGER consolidation by @glenn-jocher in LOGGER consolidation #5569
New DetectMultiBackend() class by @glenn-jocher in New DetectMultiBackend() class #5549
FROM nvcr.io/nvidia/pytorch:21.10-py3 by @glenn-jocher in FROM nvcr.io/nvidia/pytorch:21.10-py3 #5592
Add notebook_init() to utils/init.py by @glenn-jocher in Add notebook_init() to utils/__init__.py #5488
Fix check_requirements() resource warning allocation open file by @ayman-saleh in Fix check_requirements() resource warning allocation open file #5602
Update train, val tqdm to fixed width by @glenn-jocher in Update train, val tqdm to fixed width #5367
Update val.py speed and study tasks by @glenn-jocher in Update val.py speed and study tasks #5608
np.unique() sort fix for segments by @glenn-jocher in np.unique() sort fix for segments #5609
Improve plots.py robustness by @glenn-jocher in Improve plots.py robustness #5616
HUB dataset previews to JPEG by @glenn-jocher in HUB dataset previews to JPEG #5627
DDP WORLD_SIZE-safe dataloader workers by @glenn-jocher in DDP WORLD_SIZE-safe dataloader workers #5631
Default DataLoader shuffle=True for training by @werner-duvaud in Default DataLoader shuffle=True for training #5623
AutoAnchor and AutoBatch LOGGER by @glenn-jocher in AutoAnchor and AutoBatch LOGGER #5635
W&B refactor, handle exceptions, CI example by @AyushExel in W&B refactor, handle exceptions, CI example #5618
Replace 2 transpose() with 1 permute in TransformerBlock()` by @dingyiwei in Replace 2 transpose() with 1 permute in TransformerBlock()` #5645
Bump pip from 19.2 to 21.1 in /utils/google_app_engine by @dependabot in Bump pip from 19.2 to 21.1 in /utils/google_app_engine #5661
Update ci-testing.yml to Python 3.9 by @glenn-jocher in Update ci-testing.yml to Python 3.9 #5660
TFDetect dynamic anchor count assignment fix by @nrupatunga in TFDetect dynamic anchor count assignment fix #5668
Update train.py comment to 'Model attributes' by @glenn-jocher in Update train.py comment to 'Model attributes' #5670
Update export.py docstring by @glenn-jocher in Update export.py docstring #5689
NUM_THREADS leave at least 1 CPU free by @glenn-jocher in NUM_THREADS leave at least 1 CPU free #5706
Prune unused imports by @Borda in Prune unused imports #5711
Explicitly compute TP, FP in val.py by @glenn-jocher in Explicitly compute TP, FP in val.py #5727
Remove .autoshape() method by @glenn-jocher in Remove .autoshape() method #5694
SECURITY.md by @IL2006 in SECURITY.md #5695
Save *.npy features on detect.py --visualize by @Zengyf-CVer in Save *.npy features on detect.py --visualize #5701
Export, detect and validation with TensorRT engine file by @imyhxy in Export, detect and validation with TensorRT engine file #5699
Do not save hyp.yaml and opt.yaml on evolve by @glenn-jocher in Do not save hyp.yaml and opt.yaml on evolve #5775
fix the path error in export.py by @miknyko in fix the path error in export.py #5778
TorchScript torch==1.7.0 Path support by @miknyko in TorchScript torch==1.7.0 Path support #5781
W&B: refactor W&B tables by @AyushExel in W&B: refactor W&B tables #5737
Scope TF imports in DetectMultiBackend() by @phodgers in Scope TF imports in DetectMultiBackend() #5792
Remove NCOLS from tqdm by @glenn-jocher in Remove NCOLS from tqdm #5804
Refactor new model.warmup() method by @glenn-jocher in Refactor new model.warmup() method #5810
GCP VM from Image example by @glenn-jocher in GCP VM from Image example #5814
Bump actions/cache from 2.1.6 to 2.1.7 by @dependabot in Bump actions/cache from 2.1.6 to 2.1.7 #5816
Update dataset_stats() to cv2.INTER_AREA by @glenn-jocher in Update dataset_stats() to cv2.INTER_AREA #5821
Fix TensorRT potential unordered binding addresses by @imyhxy in Fix TensorRT potential unordered binding addresses #5826
Handle non-TTY wandb.errors.UsageError by @glenn-jocher in Handle non-TTY wandb.errors.UsageError #5839
Avoid inplace modifyingimgs in LoadStreams by @passerbythesun in Avoid inplace modifyingimgs in LoadStreams #5850
Update LoadImages ret_val=False handling by @gmt710 in Update LoadImages ret_val=False handling #5852
Update val.py by @pradeep-vishnu in Update val.py #5838
Update TorchScript suffix to *.torchscript by @glenn-jocher in Update TorchScript suffix to *.torchscript #5856
Add --workers 8 argument to val.py by @iumyx2612 in Add --workers 8 argument to val.py #5857
Update plot_lr_scheduler() by @daikankan in Update plot_lr_scheduler() #5864
Update nl after cutout() by @glenn-jocher in Update nl after cutout() #5873
AutoShape() models as DetectMultiBackend() instances by @glenn-jocher in AutoShape() models as DetectMultiBackend() instances #5845
Single-command multiple-model export by @glenn-jocher in Single-command multiple-model export #5882
Detections().tolist() explicit argument fix by @lizeng614 in Detections().tolist() explicit argument fix #5907
W&B: Fix bug in upload dataset module by @AyushExel in W&B: Fix bug in upload dataset module #5908
Add *.engine (TensorRT extensions) to .gitignore by @greg2451 in Add *.engine (TensorRT extensions) to .gitignore #5911
Add ONNX inference providers by @glenn-jocher in Add ONNX inference providers #5918
Add hardware checks to notebook_init() by @glenn-jocher in Add hardware checks to notebook_init() #5919
Revert "Update plot_lr_scheduler()" by @glenn-jocher in Revert "Update plot_lr_scheduler()" #5920
Absolute '/content/sample_data' by @glenn-jocher in Absolute '/content/sample_data' #5922
Default PyTorch Hub to autocast(False) by @glenn-jocher in Default PyTorch Hub to autocast(False) #5926
Fix ONNX opset inconsistency with parseargs and run args by @d57montes in Fix ONNX opset inconsistency with parseargs and run args #5937
Make select_device() robust to batch_size=-1 by @youyuxiansen in Make select_device() robust to batch_size=-1 #5940
fix .gitignore not tracking existing folders by @pasmai in fix .gitignore not tracking existing folders #5946
Update strip_optimizer() by @iumyx2612 in Update strip_optimizer() #5949
Add nms and agnostic nms to export.py by @d57montes in Add nms and agnostic nms to export.py #5938
Refactor NUM_THREADS by @glenn-jocher in Refactor NUM_THREADS #5954
Fix Detections class tolist() method by @yonomitt in Fix Detections class tolist() method #5945
Fix imgsz bug by @d57montes in Fix imgsz bug #5948
pretrained=False fix by @glenn-jocher in pretrained=False fix #5966
make parameter ignore epochs by @jinmc in make parameter ignore epochs #5972
YOLOv5s6 params and FLOPs fix by @glenn-jocher in YOLOv5s6 params and FLOPs fix #5977
Update callbacks.py with __init__() by @glenn-jocher in Update callbacks.py with __init__() #5979
Increase ar_thr from 20 to 100 for better detection on slender (high aspect ratio) objects by @MrinalJain17 in Increase ar_thr from 20 to 100 for better detection on slender (high aspect ratio) objects #5556
Allow --weights URL by @glenn-jocher in Allow --weights URL #5991
Recommend jar xf file.zip for zips by @glenn-jocher in Recommend jar xf file.zip for zips #5993
*.torchscript inference self.jit fix by @glenn-jocher in *.torchscript inference self.jit fix #6007
Check TensorRT>=8.0.0 version by @glenn-jocher in Check TensorRT>=8.0.0 version #6021
Multi-layer capable --freeze argument by @youyuxiansen in Multi-layer capable --freeze argument #6019
train -> val comment fix by @glenn-jocher in train -> val comment fix #6024
Add dataset source citations by @glenn-jocher in Add dataset source citations #6032
Kaggle LOGGER fix by @glenn-jocher in Kaggle LOGGER fix #6041
Simplify set_logging() indexing by @glenn-jocher in Simplify set_logging() indexing #6042
--freeze fix by @glenn-jocher in --freeze fix #6044
OpenVINO Export by @glenn-jocher in OpenVINO Export #6057
Reduce G/D/CIoU logic operations by @jedi007 in Reduce G/D/CIoU logic operations #6074
Init tensor directly on device by @deepsworld in Init tensor directly on device #6068
W&B: track batch size after autobatch by @AyushExel in W&B: track batch size after autobatch #6039
W&B: Log best results after training ends by @AyushExel in W&B: Log best results after training ends #6120
Log best results by @awsaf49 in Log best results #6085
Refactor/reduce G/C/D/IoU if: else statements by @cmoseses in Refactor/reduce G/C/D/IoU if: else statements #6087
Add EdgeTPU support by @zldrobit in Add EdgeTPU support #3630
Enable AdamW optimizer by @bilzard in Enable AdamW optimizer #6152
Update export format docstrings by @glenn-jocher in Update export format docstrings #6151
Update greetings.yml by @glenn-jocher in Update greetings.yml #6165
[pre-commit.ci] pre-commit suggestions by @pre-commit-ci in [pre-commit.ci] pre-commit suggestions #6177
Update NMS max_wh=7680 for 8k images by @glenn-jocher in Update NMS max_wh=7680 for 8k images #6178
Add OpenVINO inference by @glenn-jocher in Add OpenVINO inference #6179
Ignore *_openvino_model/ dir by @glenn-jocher in Ignore *_openvino_model/ dir #6180
Global export format sort by @glenn-jocher in Global export format sort #6182
Fix TorchScript on mobile export by @yinrong in Fix TorchScript on mobile export #6183
TensorRT 7 anchor_grid compatibility fix by @imyhxy in TensorRT 7 anchor_grid compatibility fix #6185
Add tensorrt>=7.0.0 checks by @glenn-jocher in Add tensorrt>=7.0.0 checks #6193
Add CoreML inference by @glenn-jocher in Add CoreML inference #6195
Fix nan-robust stream FPS by @glenn-jocher in Fix nan-robust stream FPS #6198
Edge TPU compiler comment by @glenn-jocher in Edge TPU compiler comment #6196
TFLite --int8 'flatbuffers==1.12' fix by @glenn-jocher in TFLite --int8 'flatbuffers==1.12' fix #6216
TFLite --int8 'flatbuffers==1.12' fix 2 by @glenn-jocher in TFLite --int8 'flatbuffers==1.12' fix 2 #6217
Add edgetpu_compiler checks by @glenn-jocher in Add edgetpu_compiler checks #6218
Attempt edgetpu-compiler autoinstall by @glenn-jocher in Attempt edgetpu-compiler autoinstall #6223
Update README speed reproduction command by @glenn-jocher in Update README speed reproduction command #6228
Update P2-P7 models/hub variants by @glenn-jocher in Update P2-P7 models/hub variants #6230
TensorRT 7 export fix by @glenn-jocher in TensorRT 7 export fix #6235
Fix cmd string on tfjs export by @dart-bird in Fix cmd string on tfjs export #6243
Enable ONNX --half FP16 inference by @glenn-jocher in Enable ONNX --half FP16 inference #6268
Update export.py with Detect, Validate usages by @glenn-jocher in Update export.py with Detect, Validate usages #6280
Add is_kaggle() function by @glenn-jocher in Add is_kaggle() function #6285
Fix device count check by @glenn-jocher in Fix device count check #6290
Fixing bug multi-gpu training by @hdnh2006 in Fixing bug multi-gpu training #6299
select_device() cleanup by @glenn-jocher in select_device() cleanup #6302
Fix train.py parameter groups desc error by @Otfot in Fix train.py parameter groups desc error #6318
Remove dataset_stats() autodownload capability by @glenn-jocher in Remove dataset_stats() autodownload capability #6303
Console corrupted -> corrupt by @glenn-jocher in Console corrupted -> corrupt #6338
TensorRT assert im.device.type != 'cpu' on export by @glenn-jocher in TensorRT assert im.device.type != 'cpu' on export #6340
export.py return exported files/dirs by @glenn-jocher in export.py return exported files/dirs #6343
export.py automatic forward_export by @glenn-jocher in export.py automatic forward_export #6352
Add optional VERBOSE environment variable by @johnk2hawaii in Add optional VERBOSE environment variable #6353
Reuse de_parallel() rather than is_parallel() by @imyhxy in Reuse de_parallel() rather than is_parallel() #6354
DEVICE_COUNT instead of WORLD_SIZE to calculate nw by @sitecao in DEVICE_COUNT instead of WORLD_SIZE to calculate nw #6324
Flush callbacks when on --evolve by @AyushExel in Flush callbacks when on --evolve #6374
FROM nvcr.io/nvidia/pytorch:21.12-py3 by @glenn-jocher in FROM nvcr.io/nvidia/pytorch:21.12-py3 #6377
FROM nvcr.io/nvidia/pytorch:21.10-py3 by @glenn-jocher in FROM nvcr.io/nvidia/pytorch:21.10-py3 #6379
Add albumentations to Dockerfile by @glenn-jocher in Add albumentations to Dockerfile #6392
Add stop_training=False flag to callbacks by @haimat in Add stop_training=False flag to callbacks #6365
Add detect.py GIF video inference by @glenn-jocher in Add detect.py GIF video inference #6410
Update greetings.yaml email address by @glenn-jocher in Update greetings.yaml email address #6412
Rename logger from 'utils.logger' to 'yolov5' by @JonathanSamelson in Rename logger from 'utils.logger' to 'yolov5' #6421
Prefer tflite_runtime for TFLite inference if installed by @motokimura in Prefer tflite_runtime for TFLite inference if installed #6406
Update workflows by @glenn-jocher in Update workflows #6427
Namespace VERBOSE env variable to YOLOv5_VERBOSE by @glenn-jocher in Namespace VERBOSE env variable to YOLOv5_VERBOSE #6428
Add *.asf video support by @toschi23 in Add *.asf video support #6436
Revert "Remove dataset_stats() autodownload capability" by @glenn-jocher in Revert "Remove dataset_stats() autodownload capability" #6442
Fix select_device() for Multi-GPU by @glenn-jocher in Fix select_device() for Multi-GPU #6434
Fix2 select_device() for Multi-GPU by @glenn-jocher in Fix2 select_device() for Multi-GPU #6461
Add Product Hunt social media icon by @glenn-jocher in Add Product Hunt social media icon #6464
Resolve dataset paths by @glenn-jocher in Resolve dataset paths #6489
Simplify TF normalized to pixels by @glenn-jocher in Simplify TF normalized to pixels #6494
Improved export.py usage examples by @glenn-jocher in Improved export.py usage examples #6495
CoreML inference fix list() -> sorted() by @glenn-jocher in CoreML inference fix list() -> sorted() #6496
Suppress torch.jit.TracerWarning on export by @glenn-jocher in Suppress torch.jit.TracerWarning on export #6498
Suppress export.run() TracerWarning by @glenn-jocher in Suppress export.run() TracerWarning #6499
W&B: Remember batch_size on resuming by @AyushExel in W&B: Remember batch_size on resuming #6512
Update hyp.scratch-high.yaml lrf: 0.1 by @glenn-jocher in Update hyp.scratch-high.yaml lrf: 0.1 #6525
TODO issues exempt from stale action by @glenn-jocher in TODO issues exempt from stale action #6530
Update val_batch*.jpg for Chinese fonts by @glenn-jocher in Update val_batch*.jpg for Chinese fonts #6526
Social icons after text by @glenn-jocher in Social icons after text #6473
Edge TPU compiler sudo fix by @glenn-jocher in Edge TPU compiler sudo fix #6531
Edge TPU export 'list index out of range' fix by @glenn-jocher in Edge TPU export 'list index out of range' fix #6533
Edge TPU tf.lite.experimental.load_delegate fix by @glenn-jocher in Edge TPU tf.lite.experimental.load_delegate fix #6536
Fixing minor multi-streaming issues with TensoRT engine by @greg2451 in Fixing minor multi-streaming issues with TensoRT engine #6504
Load checkpoint on CPU instead of on GPU by @bilzard in Load checkpoint on CPU instead of on GPU #6516
flake8: code meanings by @Borda in flake8: code meanings #6481
Fix 6 Flake8 issues by @Borda in Fix 6 Flake8 issues #6541
Edge TPU TF imports fix by @glenn-jocher in Edge TPU TF imports fix #6542
Move trainloader functions to class methods by @glenn-jocher in Move trainloader functions to class methods #6559
Improved AutoBatch DDP error message by @glenn-jocher in Improved AutoBatch DDP error message #6568
Fix zero-export handling with if any(f): by @glenn-jocher in Fix zero-export handling with if any(f): #6569
Fix plot_labels() colored histogram bug by @glenn-jocher in Fix plot_labels() colored histogram bug #6574
Allow custom --evolve project names by @MattVAD in Allow custom --evolve project names #6567
Add DATASETS_DIR global in general.py by @glenn-jocher in Add DATASETS_DIR global in general.py #6578
return opt from train.run() by @chf4850 in return opt from train.run() #6581
Fix YouTube dislike button bug in pafy package by @glenn-jocher in Fix YouTube dislike button bug in pafy package #6603
Fix hyp_evolve.yaml indexing bug by @glenn-jocher in Fix hyp_evolve.yaml indexing bug #6604
Fix ROOT / data when running W&B log_dataset() by @or-toledano in Fix ROOT / data when running W&B log_dataset() #6606
YouTube dependency fix youtube_dl==2020.12.2 by @glenn-jocher in YouTube dependency fix youtube_dl==2020.12.2 #6612
Add YOLOv5n to Reproduce section by @glenn-jocher in Add YOLOv5n to Reproduce section #6619
W&B: Improve resume stability by @AyushExel in W&B: Improve resume stability #6611
W&B: don't log media in evolve by @AyushExel in W&B: don't log media in evolve #6617
YOLOv5 Export Benchmarks by @glenn-jocher in YOLOv5 CPU Export Benchmarks #6613
Fix ConfusionMatrix scale vmin=0.0 by @glenn-jocher in Fix ConfusionMatrix scale vmin=0.0 #6638
Fixed wandb logger KeyError by @imyhxy in Fixed wandb logger KeyError #6637
Fix yolov3.yaml remove list by @glenn-jocher in Fix yolov3.yaml remove list #6655
Validate with 2x --workers by @glenn-jocher in Validate with 2x --workers #6658
Validate with 2x --workers single-GPU/CPU fix by @glenn-jocher in Validate with 2x --workers single-GPU/CPU fix #6659
Add --cache val option by @glenn-jocher in Add --cache val option #6663
Robust scipy.cluster.vq.kmeans too few points by @glenn-jocher in Robust scipy.cluster.vq.kmeans too few points #6668
Update Dockerfile torch==1.10.2+cu113 by @glenn-jocher in Update Dockerfile torch==1.10.2+cu113 #6669
FROM nvcr.io/nvidia/pytorch:22.01-py3 by @glenn-jocher in FROM nvcr.io/nvidia/pytorch:22.01-py3 #6670
FROM nvcr.io/nvidia/pytorch:21.10-py3 by @glenn-jocher in FROM nvcr.io/nvidia/pytorch:21.10-py3 #6671
Update Dockerfile reorder installs by @glenn-jocher in Update Dockerfile reorder installs #6672
FROM nvcr.io/nvidia/pytorch:22.01-py3 by @glenn-jocher in FROM nvcr.io/nvidia/pytorch:22.01-py3 #6673
FROM nvcr.io/nvidia/pytorch:21.10-py3 by @glenn-jocher in FROM nvcr.io/nvidia/pytorch:21.10-py3 #6677
Fix TF exports >= 2GB by @zldrobit in Fix TF exports >= 2GB #6292
Fix --evolve --bucket gs://... by @glenn-jocher in Fix --evolve --bucket gs://... #6698
Fix CoreML P6 inference by @glenn-jocher in Fix CoreML P6 inference #6700
Fix floating point in number of workers by @SamuelYvon in Fix floating point in number of workers #6701
Edge TPU inference fix by @RaffaeleGalliera in Edge TPU inference fix #6686
Use export_formats() in export.py by @glenn-jocher in Use export_formats() in export.py #6705
Suppress torch AMP-CPU warnings by @glenn-jocher in Suppress torch AMP-CPU warnings #6706
Update nw to max(nd, 1) by @glenn-jocher in Update nw to max(nd, 1) #6714
GH: add PR template by @Borda in GH: add PR template #6482
Switch default LR scheduler from cos to linear by @glenn-jocher in Switch default LR scheduler from cos to linear #6729
Updated VOC hyperparameters by @glenn-jocher in Updated VOC hyperparameters #6732

New Contributors (48)

@YoniChechik made their first contribution in fix tf conversion in new v6 models #5153
@farleylai made their first contribution in Update Objects365.yaml to include the official validation set #5194
@jdfr made their first contribution in autosplit: take image files with uppercase extensions into account #5269
@pranathlcp made their first contribution in Fixed a small typo in CONTRIBUTING.md #5445
@MrinalJain17 made their first contribution in Update check_git_status() to run under ROOT working directory #5441
@Namburger made their first contribution in Handle edgetpu model inference #5372
@ys31jp made their first contribution in Update plots.py feature_visualization path issues #5519
@nanmi made their first contribution in Update export.py #5471
@wonbeomjang made their first contribution in Update train.py #5451
@deepsworld made their first contribution in Suppress ONNX export trace warning #5437
@ayman-saleh made their first contribution in Fix check_requirements() resource warning allocation open file #5602
@werner-duvaud made their first contribution in Default DataLoader shuffle=True for training #5623
@nrupatunga made their first contribution in TFDetect dynamic anchor count assignment fix #5668
@IL2006 made their first contribution in SECURITY.md #5695
@Zengyf-CVer made their first contribution in Save *.npy features on detect.py --visualize #5701
@miknyko made their first contribution in fix the path error in export.py #5778
@phodgers made their first contribution in Scope TF imports in DetectMultiBackend() #5792
@passerbythesun made their first contribution in Avoid inplace modifyingimgs in LoadStreams #5850
@gmt710 made their first contribution in Update LoadImages ret_val=False handling #5852
@pradeep-vishnu made their first contribution in Update val.py #5838
@iumyx2612 made their first contribution in Add --workers 8 argument to val.py #5857
@daikankan made their first contribution in Update plot_lr_scheduler() #5864
@lizeng614 made their first contribution in Detections().tolist() explicit argument fix #5907
@greg2451 made their first contribution in Add *.engine (TensorRT extensions) to .gitignore #5911
@youyuxiansen made their first contribution in Make select_device() robust to batch_size=-1 #5940
@pasmai made their first contribution in fix .gitignore not tracking existing folders #5946
@yonomitt made their first contribution in Fix Detections class tolist() method #5945
@jinmc made their first contribution in make parameter ignore epochs #5972
@jedi007 made their first contribution in Reduce G/D/CIoU logic operations #6074
@awsaf49 made their first contribution in Log best results #6085
@cmoseses made their first contribution in Refactor/reduce G/C/D/IoU if: else statements #6087
@bilzard made their first contribution in Enable AdamW optimizer #6152
@pre-commit-ci made their first contribution in [pre-commit.ci] pre-commit suggestions #6177
@yinrong made their first contribution in Fix TorchScript on mobile export #6183
@dart-bird made their first contribution in Fix cmd string on tfjs export #6243
@hdnh2006 made their first contribution in Fixing bug multi-gpu training #6299
@Otfot made their first contribution in Fix train.py parameter groups desc error #6318
@johnk2hawaii made their first contribution in Add optional VERBOSE environment variable #6353
@sitecao made their first contribution in DEVICE_COUNT instead of WORLD_SIZE to calculate nw #6324
@haimat made their first contribution in Add stop_training=False flag to callbacks #6365
@JonathanSamelson made their first contribution in Rename logger from 'utils.logger' to 'yolov5' #6421
@motokimura made their first contribution in Prefer tflite_runtime for TFLite inference if installed #6406
@toschi23 made their first contribution in Add *.asf video support #6436
@MattVAD made their first contribution in Allow custom --evolve project names #6567
@chf4850 made their first contribution in return opt from train.run() #6581
@or-toledano made their first contribution in Fix ROOT / data when running W&B log_dataset() #6606
@SamuelYvon made their first contribution in Fix floating point in number of workers #6701
@RaffaeleGalliera made their first contribution in Edge TPU inference fix #6686

Full Changelog: v6.0...v6.1

This discussion was created from the release v6.1 - TensorRT, TensorFlow Edge TPU and OpenVINO Export and Inference.

aliencaocao · 2022-02-22T12:35:20Z

aliencaocao
Feb 22, 2022

The tensorflow savedmodel link is not working for me. When I click on it, it just downloads the website instead of actually showing it. All other links there works.

2 replies

glenn-jocher Feb 22, 2022
Maintainer Author

@aliencaocao thanks for the feedback! I think its a bug on the tensorflow website, i.e. when I just google 'tensorflow saved model' and click the first link the same thing happens: https://www.google.com/search?q=tensorflow+saved+model

aliencaocao Feb 22, 2022

Yes, many other links under the same topic on tensorflow website also has the same issue. I guess have to wait for them to fix.

Charles-931 · 2022-05-06T10:43:52Z

Charles-931
May 6, 2022

Will you post the pretrained weights for YOLOv5s-P2?

2 replies

glenn-jocher May 6, 2022
Maintainer Author

@Charles-931 no, we don't provide trained models for any of the yamls in models/hub other than the P6 models. You can train one of these yourself though very easily. For the YOLOv5l P2 model the 'official' training command would be:

python train.py --data coco.yaml --batch 128 --weights --cfg yolov5-p2.yaml --epochs 300 --hyp hyp.scratch-high.yaml

glenn-jocher May 6, 2022
Maintainer Author

All official YOLOv5 🚀 pretrained models including P6 models are available to download directly under the latest release in the 'Assets' section, i.e.:
https://github.com/ultralytics/yolov5/releases/tag/v6.1

Good luck 🍀 and let us know if you have any other questions!

tino926 · 2022-05-10T08:25:25Z

tino926
May 10, 2022

Is v6.1 not the latest release now?
Today I found on the yolov5 github page that the latest release is v6.0

Does that mean v6.1 is found less stable than v6.0? I can't find any discussion about it.
I am currently using v6.1 training my model, so I really want to know what happens.

1 reply

glenn-jocher May 10, 2022
Maintainer Author

@tino926 hi thanks for letting us know! Nothing has changed, the v6.1 tag remains, but for some reason GitHub was not showing it as the latest. I think it may because I went to edit the release notes and did not save, so was left in edit mode and GitHub reverted to the latest saved. I've fixed this and everything is back to normal.

The latest master commit is always recommended as it incorporates the latest fixes and features.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

v6.1 - TensorRT, TensorFlow Edge TPU and OpenVINO Export and Inference #6740

{{title}}

Replies: 3 comments 5 replies

{{title}}

{{title}}

{{title}}

{{title}}

{{title}}

{{title}}

{{title}}

{{title}}

Select a reply

v6.1 - TensorRT, TensorFlow Edge TPU and OpenVINO Export and Inference #6740

glenn-jocher Feb 22, 2022 Maintainer

Important Updates

New Results

Pretrained Checkpoints

Changelog

Replies: 3 comments · 5 replies

aliencaocao Feb 22, 2022

glenn-jocher Feb 22, 2022 Maintainer Author

aliencaocao Feb 22, 2022

Charles-931 May 6, 2022

glenn-jocher May 6, 2022 Maintainer Author

glenn-jocher May 6, 2022 Maintainer Author

tino926 May 10, 2022

glenn-jocher May 10, 2022 Maintainer Author

glenn-jocher
Feb 22, 2022
Maintainer

Replies: 3 comments 5 replies

aliencaocao
Feb 22, 2022

glenn-jocher Feb 22, 2022
Maintainer Author

Charles-931
May 6, 2022

glenn-jocher May 6, 2022
Maintainer Author

glenn-jocher May 6, 2022
Maintainer Author

tino926
May 10, 2022

glenn-jocher May 10, 2022
Maintainer Author