Yolov5 Versions Save

YOLOv5 🚀 in PyTorch > ONNX > CoreML > TFLite

v7.0

1 year ago

Our new YOLOv5 v7.0 instance segmentation models are the fastest and most accurate in the world, beating all current SOTA benchmarks. We've made them super simple to train, validate and deploy. See full details in our Release Notes and visit our YOLOv5 Segmentation Colab Notebook for quickstart tutorials.


Our primary goal with this release is to introduce super simple YOLOv5 segmentation workflows just like our existing object detection models. The new v7.0 YOLOv5-seg models below are just a start, we will continue to improve these going forward together with our existing detection and classification models. We'd love your feedback and contributions on this effort!

This release incorporates 280 PRs from 41 contributors since our last release in August 2022.

Important Updates

New Segmentation Checkpoints

We trained YOLOv5 segmentations models on COCO for 300 epochs at image size 640 using A100 GPUs. 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 notebooks for easy reproducibility.

Model size
(pixels)		 mAPbox
50-95		 mAPmask
50-95		 Train time
300 epochs
A100 (hours)		 Speed
ONNX CPU
(ms)		 Speed
TRT A100
(ms)		 params
(M)		 FLOPs
@640 (B)
YOLOv5n-seg 640 27.6 23.4 80:17 62.7 1.2 2.0 7.1
YOLOv5s-seg 640 37.6 31.7 88:16 173.3 1.4 7.6 26.4
YOLOv5m-seg 640 45.0 37.1 108:36 427.0 2.2 22.0 70.8
YOLOv5l-seg 640 49.0 39.9 66:43 (2x) 857.4 2.9 47.9 147.7
YOLOv5x-seg 640 50.7 41.4 62:56 (3x) 1579.2 4.5 88.8 265.7
  • All checkpoints are trained to 300 epochs with SGD optimizer with lr0=0.01 and weight_decay=5e-5 at image size 640 and all default settings.
    Runs logged to https://wandb.ai/glenn-jocher/YOLOv5_v70_official
  • Accuracy values are for single-model single-scale on COCO dataset.
    Reproduce by python segment/val.py --data coco.yaml --weights yolov5s-seg.pt
  • Speed averaged over 100 inference images using a Colab Pro A100 High-RAM instance. Values indicate inference speed only (NMS adds about 1ms per image).
    Reproduce by python segment/val.py --data coco.yaml --weights yolov5s-seg.pt --batch 1
  • Export to ONNX at FP32 and TensorRT at FP16 done with export.py.
    Reproduce by python export.py --weights yolov5s-seg.pt --include engine --device 0 --half

New Segmentation Usage Examples

Train

YOLOv5 segmentation training supports auto-download COCO128-seg segmentation dataset with --data coco128-seg.yaml argument and manual download of COCO-segments dataset with bash data/scripts/get_coco.sh --train --val --segments and then python train.py --data coco.yaml.

# Single-GPU
python segment/train.py --model yolov5s-seg.pt --data coco128-seg.yaml --epochs 5 --img 640

# Multi-GPU DDP
python -m torch.distributed.run --nproc_per_node 4 --master_port 1 segment/train.py --model yolov5s-seg.pt --data coco128-seg.yaml --epochs 5 --img 640 --device 0,1,2,3

Val

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

bash data/scripts/get_coco.sh --val --segments  # download COCO val segments split (780MB, 5000 images)
python segment/val.py --weights yolov5s-seg.pt --data coco.yaml --img 640  # validate

Predict

Use pretrained YOLOv5m-seg to predict bus.jpg:

python segment/predict.py --weights yolov5m-seg.pt --data data/images/bus.jpg

model = torch.hub.load('ultralytics/yolov5', 'custom', 'yolov5m-seg.pt')  # load from PyTorch Hub (WARNING: inference not yet supported)
zidane bus

Export

Export YOLOv5s-seg model to ONNX and TensorRT:

python export.py --weights yolov5s-seg.pt --include onnx engine --img 640 --device 0

Changelog

🛠️ New Features and Bug Fixes (280) * Improve classification comments by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/8997 * Update `attempt_download(release='v6.2')` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/8998 * Update README_cn.md by @KieraMengru0907 in https://github.com/ultralytics/yolov5/pull/9001 * Update dataset `names` from array to dictionary by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9000 * [segment]: Allow inference on dirs and videos by @AyushExel in https://github.com/ultralytics/yolov5/pull/9003 * DockerHub tag update Usage example by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9005 * Add weight `decay` to argparser by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9006 * Add glob quotes to detect.py usage example by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9007 * Fix TorchScript JSON string key bug by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9015 * EMA FP32 assert classification bug fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9016 * Faster pre-processing for gray image input by @cher-liang in https://github.com/ultralytics/yolov5/pull/9009 * Improved `Profile()` inference timing by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9024 * `torch.empty()` for speed improvements by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9025 * Remove unused `time_sync` import by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9026 * Add PyTorch Hub classification CI checks by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9027 * Attach transforms to model by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9028 * Default --data `imagenette160` training (fastest) by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9033 * VOC `names` dictionary fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9034 * Update train.py `import val as validate` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9037 * AutoBatch protect from negative batch sizes by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9048 * Temporarily remove `macos-latest` from CI by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9049 * Add `--save-hybrid` mAP warning by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9050 * Refactor for simplification by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9054 * Refactor for simplification 2 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9055 * zero-mAP fix return `.detach()` to EMA by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9056 * zero-mAP fix 3 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9058 * Daemon `plot_labels()` for faster start by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9057 * TensorBoard fix in tutorial.ipynb by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9064 * zero-mAP fix remove `torch.empty()` forward pass in `.train()` mode by @0zppd in https://github.com/ultralytics/yolov5/pull/9068 * Rename 'labels' to 'instances' by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9066 * Threaded TensorBoard graph logging by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9070 * De-thread TensorBoard graph logging by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9071 * Two dimensional `size=(h,w)` AutoShape support by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9072 * Remove unused Timeout import by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9073 * Improved Usage example docstrings by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9075 * Install `torch` latest stable by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9092 * New `@try_export` decorator by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9096 * Add optional `transforms` argument to LoadStreams() by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9105 * Streaming Classification support by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9106 * Fix numpy to torch cls streaming bug by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9112 * Infer Loggers project name by @AyushExel in https://github.com/ultralytics/yolov5/pull/9117 * Add CSV logging to GenericLogger by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9128 * New TryExcept decorator by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9154 * Fixed segment offsets by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9155 * New YOLOv5 v6.2 splash images by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9142 * Rename onnx_dynamic -> dynamic by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9168 * Inline `_make_grid()` meshgrid by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9170 * Comment EMA assert by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9173 * Fix confidence threshold for ClearML debug images by @HighMans in https://github.com/ultralytics/yolov5/pull/9174 * Update Dockerfile-cpu by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9184 * Update Dockerfile-cpu to libpython3-dev by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9185 * Update Dockerfile-arm64 to libpython3-dev by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9187 * Fix AutoAnchor MPS bug by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9188 * Skip AMP check on MPS by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9189 * ClearML's set_report_period's time is defined in minutes not seconds. by @HighMans in https://github.com/ultralytics/yolov5/pull/9186 * Add `check_git_status(..., branch='master')` argument by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9199 * `check_font()` on notebook init by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9200 * Comment `protobuf` in requirements.txt by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9207 * `check_font()` fstring update by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9208 * AutoBatch protect from extreme batch sizes by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9209 * Default AutoBatch 0.8 fraction by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9212 * Delete rebase.yml by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9202 * Duplicate segment verification fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9225 * New `LetterBox(size)` `CenterCrop(size)`, `ToTensor()` transforms (#9213) by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9213 * Add ClassificationModel TF export assert by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9226 * Remove usage of `pathlib.Path.unlink(missing_ok=...)` by @ymerkli in https://github.com/ultralytics/yolov5/pull/9227 * Add support for `*.pfm` images by @spacewalk01 in https://github.com/ultralytics/yolov5/pull/9230 * Python check warning emoji by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9238 * Add `url_getsize()` function by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9247 * Update dataloaders.py by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9250 * Refactor Loggers : Move code outside train.py by @AyushExel in https://github.com/ultralytics/yolov5/pull/9241 * Update general.py by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9252 * Add LoadImages._cv2_rotate() by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9249 * Move `cudnn.benchmarks(True)` to LoadStreams by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9258 * `cudnn.benchmark = True` on Seed 0 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9259 * Update `TryExcept(msg='...')`` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9261 * Make sure best.pt model file is preserved ClearML by @thepycoder in https://github.com/ultralytics/yolov5/pull/9265 * DetectMultiBackend improvements by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9269 * Update DetectMultiBackend for tuple outputs by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9274 * Update DetectMultiBackend for tuple outputs 2 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9275 * Update benchmarks CI with `--hard-fail` min metric floor by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9276 * Add new `--vid-stride` inference parameter for videos by @VELCpro in https://github.com/ultralytics/yolov5/pull/9256 * [pre-commit.ci] pre-commit suggestions by @pre-commit-ci in https://github.com/ultralytics/yolov5/pull/9295 * Replace deprecated `np.int` with `int` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9307 * Comet Logging and Visualization Integration by @DN6 in https://github.com/ultralytics/yolov5/pull/9232 * Comet changes by @DN6 in https://github.com/ultralytics/yolov5/pull/9328 * Train.py line 486 typo fix by @robinned in https://github.com/ultralytics/yolov5/pull/9330 * Add dilated conv support by @YellowAndGreen in https://github.com/ultralytics/yolov5/pull/9347 * Update `check_requirements()` single install by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9353 * Update `check_requirements(args, cmds='')` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9355 * Update `check_requirements()` multiple string by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9356 * Add PaddlePaddle export and inference by @kisaragychihaya in https://github.com/ultralytics/yolov5/pull/9240 * PaddlePaddle Usage examples by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9358 * labels.jpg names fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9361 * Exclude `ipython` from hubconf.py `check_requirements()` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9362 * `torch.jit.trace()` fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9363 * AMP Check fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9367 * Remove duplicate line in setup.cfg by @zldrobit in https://github.com/ultralytics/yolov5/pull/9380 * Remove `.train()` mode exports by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9429 * Continue on Docker arm64 failure by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9430 * Continue on Docker failure (all backends) by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9432 * Continue on Docker fail (all backends) fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9433 * YOLOv5 segmentation model support by @AyushExel in https://github.com/ultralytics/yolov5/pull/9052 * Fix val.py zero-TP bug by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9431 * New model.yaml `activation:` field by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9371 * Fix tick labels for background FN/FP by @hotohoto in https://github.com/ultralytics/yolov5/pull/9414 * Fix TensorRT exports to ONNX opset 12 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9441 * AutoShape explicit arguments fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9443 * Update Detections() instance printing by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9445 * AutoUpdate TensorFlow in export.py by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9447 * AutoBatch `cudnn.benchmark=True` fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9448 * Do not move downloaded zips by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9455 * Update general.py by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9454 * `Detect()` and `Segment()` fixes for CoreML and Paddle by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9458 * Add Paddle exports to benchmarks by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9459 * Add `macos-latest` runner for CoreML benchmarks by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9453 * Fix cutout bug by @Oswells in https://github.com/ultralytics/yolov5/pull/9452 * Optimize imports by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9464 * TensorRT SegmentationModel fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9465 * `Conv()` dilation argument fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9466 * Update ClassificationModel default training `imgsz=224` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9469 * Standardize warnings with `WARNING ⚠️ ...` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9467 * TensorFlow macOS AutoUpdate by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9471 * `segment/predict --save-txt` fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9478 * TensorFlow SegmentationModel support by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9472 * AutoBatch report include reserved+allocated by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9491 * Update Detect() grid init `for` loop by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9494 * Accelerate video inference by @mucunwuxian in https://github.com/ultralytics/yolov5/pull/9487 * Comet Image Logging Fix by @DN6 in https://github.com/ultralytics/yolov5/pull/9498 * Fix visualization title bug by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9500 * Add paddle tips by @Zengyf-CVer in https://github.com/ultralytics/yolov5/pull/9502 * Segmentation `polygons2masks_overlap()` in `np.int32` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9493 * Fix `random_perspective` param bug in segment by @FeiGeChuanShu in https://github.com/ultralytics/yolov5/pull/9512 * Remove `check_requirements('flatbuffers==1.12')` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9514 * Fix TF Lite exports by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9517 * TFLite fix 2 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9518 * FROM nvcr.io/nvidia/pytorch:22.08-py3 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9520 * Remove scikit-learn constraint on coremltools 6.0 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9530 * Update scikit-learn constraint per coremltools 6.0 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9531 * Update `coremltools>=6.0` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9532 * Update albumentations by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9503 * import re by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9535 * TF.js fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9536 * Refactor dataset batch-size by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9551 * Add `--source screen` for screenshot inference by @zombob in https://github.com/ultralytics/yolov5/pull/9542 * Update `is_url()` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9566 * Detect.py supports running against a Triton container by @gaziqbal in https://github.com/ultralytics/yolov5/pull/9228 * New `scale_segments()` function by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9570 * generator seed fix for DDP mAP drop by @Forever518 in https://github.com/ultralytics/yolov5/pull/9545 * Update default GitHub assets by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9573 * Update requirements.txt comment https://pytorch.org/get-started/locally/ by @davidamacey in https://github.com/ultralytics/yolov5/pull/9576 * Add segment line predictions by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9571 * TensorRT detect.py inference fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9581 * Update Comet links by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9587 * Add global YOLOv5_DATASETS_DIR by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9586 * Add Paperspace Gradient badges by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9588 * #YOLOVISION22 announcement by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9590 * Bump actions/stale from 5 to 6 by @dependabot in https://github.com/ultralytics/yolov5/pull/9595 * #YOLOVISION22 update by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9598 * Apple MPS -> CPU NMS fallback strategy by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9600 * Updated Segmentation and Classification usage by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9607 * Update export.py Usage examples by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9609 * Fix `is_url('https://ultralytics.com')` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9610 * Add `results.save(save_dir='path', exist_ok=False)` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9617 * NMS MPS device wrapper by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9620 * Add SegmentationModel unsupported warning by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9632 * Disabled upload_dataset flag temporarily due to an artifact related bug by @soumik12345 in https://github.com/ultralytics/yolov5/pull/9652 * Add NVIDIA Jetson Nano Deployment tutorial by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9656 * Added cutout import from utils/augmentations.py to use Cutout Aug in … by @senhorinfinito in https://github.com/ultralytics/yolov5/pull/9668 * Simplify val.py benchmark mode with speed mode by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9674 * Allow list for Comet artifact class 'names' field by @KristenKehrer in https://github.com/ultralytics/yolov5/pull/9654 * [pre-commit.ci] pre-commit suggestions by @pre-commit-ci in https://github.com/ultralytics/yolov5/pull/9685 * TensorRT `--dynamic` fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9691 * FROM nvcr.io/nvidia/pytorch:22.09-py3 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9711 * Error in utils/segment/general `masks2segments()` by @paulguerrie in https://github.com/ultralytics/yolov5/pull/9724 * Fix segment evolution keys by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9742 * Remove #YOLOVISION22 notice by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9751 * Update Loggers by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9760 * update mask2segments and saving results by @vladoossss in https://github.com/ultralytics/yolov5/pull/9785 * HUB VOC fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9792 * Update hubconf.py local repo Usage example by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9803 * Fix xView dataloaders import by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9807 * Argoverse HUB fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9809 * `smart_optimizer()` revert to weight with decay by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9817 * Allow PyTorch Hub results to display in notebooks by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9825 * Logger Cleanup by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9828 * Remove ipython from `check_requirements` exclude list by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9841 * Update HUBDatasetStats() usage examples by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9842 * Update ZipFile to context manager by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9843 * Update README.md by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9846 * Webcam show fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9847 * Fix OpenVINO Usage example by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9874 * ClearML Dockerfile fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9876 * Windows Python 3.7 .isfile() fix by @SSTato in https://github.com/ultralytics/yolov5/pull/9879 * Add TFLite Metadata to TFLite and Edge TPU models by @paradigmn in https://github.com/ultralytics/yolov5/pull/9903 * Add `gnupg` to Dockerfile-cpu by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9932 * Add ClearML minimum version requirement by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9933 * Update Comet Integrations table text by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9937 * Update README.md by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9957 * Update README.md by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9958 * Update README.md by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9961 * Switch from suffix checks to archive checks by @kalenmike in https://github.com/ultralytics/yolov5/pull/9963 * FROM nvcr.io/nvidia/pytorch:22.10-py3 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9966 * Full-size proto code (optional) by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9980 * Update README.md by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9970 * Segmentation Tutorial by @paulguerrie in https://github.com/ultralytics/yolov5/pull/9521 * Fix `is_colab()` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9994 * Check online twice on AutoUpdate by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9999 * Add `min_items` filter option by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9997 * Improved `check_online()` robustness by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10000 * Fix `min_items` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10001 * Update default `--epochs 100` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10024 * YOLOv5 AutoCache Update by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10027 * IoU `eps` adjustment by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10051 * Update get_coco.sh by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10057 * [pre-commit.ci] pre-commit suggestions by @pre-commit-ci in https://github.com/ultralytics/yolov5/pull/10068 * Use MNIST160 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10069 * Update Dockerfile keep default torch installation by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10071 * Add `ultralytics` pip package by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10103 * AutoShape integer image-size fix by @janus-zheng in https://github.com/ultralytics/yolov5/pull/10090 * YouTube Usage example comments by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10106 * Mapped project and name to ClearML by @thepycoder in https://github.com/ultralytics/yolov5/pull/10100 * Update IoU functions by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10123 * Add Ultralytics HUB to README by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10070 * Fix benchmark.py usage comment by @rusamentiaga in https://github.com/ultralytics/yolov5/pull/10131 * Update HUB banner image by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10134 * Copy-Paste zero value fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10152 * Add Copy-Paste to `mosaic9()` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10165 * Add `join_threads()` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10086 * Fix dataloader filepath modification to perform replace only once and not for all occurences of string by @adumrewal in https://github.com/ultralytics/yolov5/pull/10163 * fix: prevent logging config clobbering by @rkechols in https://github.com/ultralytics/yolov5/pull/10133 * Filter PyTorch 1.13 UserWarnings by @triple-Mu in https://github.com/ultralytics/yolov5/pull/10166 * Revert "fix: prevent logging config clobbering" by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10177 * Apply make_divisible for ONNX models in Autoshape by @janus-zheng in https://github.com/ultralytics/yolov5/pull/10172 * data.yaml `names.keys()` integer assert by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10190 * fix: try 2 - prevent logging config clobbering by @rkechols in https://github.com/ultralytics/yolov5/pull/10192 * Segment prediction labels normalization fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10205 * Simplify dataloader tqdm descriptions by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10210 * New global `TQDM_BAR_FORMAT` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10211 * Feature/classification tutorial refactor by @paulguerrie in https://github.com/ultralytics/yolov5/pull/10039 * Remove Colab notebook High-Memory notices by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10212 * Revert `--save-txt` to default False by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10213 * Add `--source screen` Usage example by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10215 * Add `git` info to training checkpoints by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/9655 * Add git info to cls, seg checkpoints by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10217 * Update Comet preview image by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10220 * Scope gitpyhon import in `check_git_info()` by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10221 * Squeezenet reshape outputs fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/10222
😃 New Contributors (30) * @KieraMengru0907 made their first contribution in https://github.com/ultralytics/yolov5/pull/9001 * @cher-liang made their first contribution in https://github.com/ultralytics/yolov5/pull/9009 * @0zppd made their first contribution in https://github.com/ultralytics/yolov5/pull/9068 * @HighMans made their first contribution in https://github.com/ultralytics/yolov5/pull/9174 * @ymerkli made their first contribution in https://github.com/ultralytics/yolov5/pull/9227 * @spacewalk01 made their first contribution in https://github.com/ultralytics/yolov5/pull/9230 * @VELCpro made their first contribution in https://github.com/ultralytics/yolov5/pull/9256 * @DN6 made their first contribution in https://github.com/ultralytics/yolov5/pull/9232 * @robinned made their first contribution in https://github.com/ultralytics/yolov5/pull/9330 * @kisaragychihaya made their first contribution in https://github.com/ultralytics/yolov5/pull/9240 * @hotohoto made their first contribution in https://github.com/ultralytics/yolov5/pull/9414 * @Oswells made their first contribution in https://github.com/ultralytics/yolov5/pull/9452 * @mucunwuxian made their first contribution in https://github.com/ultralytics/yolov5/pull/9487 * @FeiGeChuanShu made their first contribution in https://github.com/ultralytics/yolov5/pull/9512 * @zombob made their first contribution in https://github.com/ultralytics/yolov5/pull/9542 * @gaziqbal made their first contribution in https://github.com/ultralytics/yolov5/pull/9228 * @Forever518 made their first contribution in https://github.com/ultralytics/yolov5/pull/9545 * @davidamacey made their first contribution in https://github.com/ultralytics/yolov5/pull/9576 * @soumik12345 made their first contribution in https://github.com/ultralytics/yolov5/pull/9652 * @senhorinfinito made their first contribution in https://github.com/ultralytics/yolov5/pull/9668 * @KristenKehrer made their first contribution in https://github.com/ultralytics/yolov5/pull/9654 * @paulguerrie made their first contribution in https://github.com/ultralytics/yolov5/pull/9724 * @vladoossss made their first contribution in https://github.com/ultralytics/yolov5/pull/9785 * @SSTato made their first contribution in https://github.com/ultralytics/yolov5/pull/9879 * @janus-zheng made their first contribution in https://github.com/ultralytics/yolov5/pull/10090 * @rusamentiaga made their first contribution in https://github.com/ultralytics/yolov5/pull/10131 * @adumrewal made their first contribution in https://github.com/ultralytics/yolov5/pull/10163 * @rkechols made their first contribution in https://github.com/ultralytics/yolov5/pull/10133 * @triple-Mu made their first contribution in https://github.com/ultralytics/yolov5/pull/10166

v6.2

1 year ago

v6.1

2 years ago

This release incorporates many 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

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 [email protected]: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)		 mAPval
0.5:0.95		 mAPval
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 67.3 430.0 10.1 2.7 46.5 109.1

Pretrained Checkpoints

Model size
(pixels)		 mAPval
0.5:0.95		 mAPval
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 12.6 16.8
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.
  • mAPval 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: https://github.com/ultralytics/yolov5/compare/v6.0...v6.1 Changes since this release: https://github.com/ultralytics/yolov5/compare/v6.1...HEAD

New Features and Bug Fixes (271)
New Contributors (48)

Full Changelog: https://github.com/ultralytics/yolov5/compare/v6.0...v6.1

v6.0

2 years ago

v5.0

3 years ago

This release implements YOLOv5-P6 models and retrained YOLOv5-P5 models. All model sizes YOLOv5s/m/l/x are now available in both P5 and P6 architectures:

  • YOLOv5-P5 models (same architecture as v4.0 release): 3 output layers P3, P4, P5 at strides 8, 16, 32, trained at --img 640 
python detect.py --weights yolov5s.pt  # P5 models
                           yolov5m.pt
                           yolov5l.pt
                           yolov5x.pt
  • YOLOv5-P6 models: 4 output layers P3, P4, P5, P6 at strides 8, 16, 32, 64 trained at --img 1280 
python detect.py --weights yolov5s6.pt  # P6 models
                           yolov5m6.pt
                           yolov5l6.pt
                           yolov5x6.pt

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

Notable Updates

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.
  • Reproduce by python test.py --task study --data coco.yaml --iou 0.7 --weights yolov5s6.pt yolov5m6.pt yolov5l6.pt yolov5x6.pt

Pretrained Checkpoints

Model size
(pixels)		 mAPval
0.5:0.95		 mAPtest
0.5:0.95		 mAPval
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 63.1 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)
  • APtest 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
  • SpeedGPU 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: https://github.com/ultralytics/yolov5/compare/v4.0...v5.0 Changes since this release: https://github.com/ultralytics/yolov5/compare/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)

v4.0

3 years ago

This release implements two architecture changes to YOLOv5, as well as various bug fixes and performance improvements.

Breaking Changes

  • nn.SiLU() activations replace nn.LeakyReLU(0.1) and nn.Hardswish() activations used in previous versions. nn.SiLU() was introduced in PyTorch 1.7.0 (https://pytorch.org/docs/stable/generated/torch.nn.SiLU.html), and due to the recent timeframe certain export pipelines may be temporarily unavailable (CoreML possibly) without updates to the associated tools (i.e. coremltools).

Bug Fixes

  • Multi-GPU --resume #1810
  • leaf Variable inplace bug fix #1759
  • Various additional bug fixes contained in PRs #1235 through #1837

Added Functionality

  • Weights & Biases (W&B) Feature Addition #1235
  • Utils reorganization #1392
  • PyTorch Hub and autoShape update #1415
  • W&B artifacts feature addition #1712
  • Various additional feature additions contained in PRs #1235 through #1837

Updated Results

Latest models are all slightly smaller to due removal of one convolution within each bottleneck, which have been renamed as C3() modules now in light of the 3 I/O convolutions each one does vs the 4 in the standard CSP bottleneck. The previous manual concatenation and LeakyReLU(0.1) activations have both removed, simplifying the architecture, reducing parameter count, and better exploiting the .fuse() operation at inference time.

nn.SiLU() activations replace nn.LeakyReLU(0.1) and nn.Hardswish() activations throughout the model, simplifying the architecture as we now only have one single activation function used everywhere rather than the two types before.

In general the changes result in smaller models (89.0M params -> 87.7M YOLOv5x), faster inference times (6.9ms -> 6.0ms), and improved mAP (49.2 -> 50.1) for all models except YOLOv5s, which reduced mAP slightly (37.0 -> 36.8). In general the largest models benefit the most from this update. YOLOv5x in particular is now above 50.0 mAP at --img-size 640, which may be the first time this is possible at 640 resolution for any architecture I'm aware of (correct me if I'm wrong though).

** 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.

Pretrained Checkpoints

Model size APval APtest AP50 SpeedV100 FPSV100 params GFLOPS
YOLOv5s 640 36.8 36.8 55.6 2.2ms 455 7.3M 17.0
YOLOv5m 640 44.5 44.5 63.1 2.9ms 345 21.4M 51.3
YOLOv5l 640 48.1 48.1 66.4 3.8ms 264 47.0M 115.4
YOLOv5x 640 50.1 50.1 68.7 6.0ms 167 87.7M 218.8
YOLOv5x + TTA 832 51.9 51.9 69.6 24.9ms 40 87.7M 1005.3

v3.1

3 years ago

This release aggregates various minor bug fixes and performance improvements since the main v3.0 release and incorporates PyTorch 1.7.0 compatibility updates. v3.1 models share weights with v3.0 models but contain minor module updates (inplace fields for nn.Hardswish() activations) for native PyTorch 1.7.0 compatibility. For PyTorch 1.7.0 release updates see https://github.com/pytorch/pytorch/releases/tag/v1.7.0.

Breaking Changes

Bug Fixes

Added Functionality

v3.0

3 years ago

This releases includes nn.Hardswish() activation implementation on Conv() modules, which increases mAP for all models at the expense of about 10% in inference speed. Training speeds are not significantly affected, though CUDA memory requirements increase about 10%. Training from scratch as well as finetuning both benefit from this change. The smallest models benefit the most from the Hardswish() activations, with increases of +0.9/+0.8/+0.7/[email protected]:0.95 for YOLOv5s/m/l/x.

All mAP values in our README are now reported at --img-size 640 (v2.0 reported at 672, and v1.0 reported at 736), so we've succeeded in increasing mAP while reducing the required --img-size :)

We've also listed YOLOv5x Test Time Augmentation (TTA) mAP and speeds for v3.0 in our README table for the first time (and for v2.0 below). Best results are YOLOv5x with TTA at 50.8 [email protected]:0.95. We've also updated efficientdet results in our comparison plot to reflect recent improvements in the google/automl repo.

Breaking Changes

  • This release does not contain breaking changes.
  • This release is only backwards compatible with v2.0 models trained with torch>=1.6.

Bug Fixes

Added Functionality

Pretrained Checkpoints

v3.0 with nn.Hardswish()

Model APval APtest AP50 SpeedGPU FPSGPU params FLOPS
YOLOv5s 37.0 37.0 56.2 2.4ms 476 7.5M 13.2B
YOLOv5m 44.3 44.3 63.2 3.4ms 333 21.8M 39.4B
YOLOv5l 47.7 47.7 66.5 4.4ms 256 47.8M 88.1B
YOLOv5x 49.2 49.2 67.7 6.9ms 164 89.0M 166.4B
YOLOv5x + TTA 50.8 50.8 68.9 25.5ms 39 89.0M 354.3B
YOLOv3-SPP 45.6 45.5 65.2 4.5ms 222 63.0M 118.0B

** APtest denotes COCO test-dev2017 server results, all other AP results in the table denote val2017 accuracy.
** All AP numbers are for single-model single-scale without ensemble or test-time augmentation except for TTA. Reproduce by python test.py --data coco.yaml --img 640 --conf 0.001. Test Time Augmentation (TTA) runs at 3 image sizes. Reproduce TTA results by python test.py --data coco.yaml --img 832 --augment ** SpeedGPU measures end-to-end time per image averaged over 5000 COCO val2017 images using a GCP n1-standard-16 instance with one V100 GPU, and includes image preprocessing, PyTorch FP16 image inference at --batch-size 32 --img-size 640, postprocessing and NMS. Average NMS time included in this chart is 1-2ms/img. Reproduce by python test.py --data coco.yaml --img 640 --conf 0.1
** All checkpoints are trained to 300 epochs with default settings and hyperparameters (no autoaugmentation).

v2.0 with nn.LeakyReLU(0.1)

Model APval APtest AP50 SpeedGPU FPSGPU params FLOPS
YOLOv5s 36.1 36.1 55.3 2.2ms 476 7.5M 13.2B
YOLOv5m 43.5 43.5 62.5 3.2ms 333 21.8M 39.4B
YOLOv5l 47.0 47.1 65.6 4.1ms 256 47.8M 88.1B
YOLOv5x 49.0 49.0 67.4 6.4ms 164 89.0M 166.4B
YOLOv5x + TTA 50.4 50.4 68.5 23.4ms 43 89.0M 354.3B
YOLOv3-SPP 45.6 45.5 65.2 4.5ms 222 63.0M 118.0B

** APtest denotes COCO test-dev2017 server results, all other AP results in the table denote val2017 accuracy.
** All AP numbers are for single-model single-scale without ensemble or test-time augmentation. Reproduce by python test.py --data coco.yaml --img 672 --conf 0.001
** SpeedGPU measures end-to-end time per image averaged over 5000 COCO val2017 images using a GCP n1-standard-16 instance with one V100 GPU, and includes image preprocessing, PyTorch FP16 image inference at --batch-size 32 --img-size 640, postprocessing and NMS. Average NMS time included in this chart is 1-2ms/img. Reproduce by python test.py --data coco.yaml --img 640 --conf 0.1
** All checkpoints are trained to 300 epochs with default settings and hyperparameters (no autoaugmentation).

v2.0

3 years ago

Breaking Changes

IMPORTANT: v2.0 release contains breaking changes. Models trained with earlier versions will not operate correctly with v2.0. The last commit before v2.0 that operates correctly with all earlier pretrained models is: https://github.com/ultralytics/yolov5/tree/5e970d45c44fff11d1eb29bfc21bed9553abf986

To clone last commit prior to v2.0:

git clone https://github.com/ultralytics/yolov5  # clone repo
cd yolov5
git reset --hard 5e970d4  # last commit before v2.0

Bug Fixes

  • Various

Added Functionality

  • Various
** GPU Speed measures end-to-end time per image averaged over 5000 COCO val2017 images using a V100 GPU with batch size 8, and includes image preprocessing, PyTorch FP16 inference, postprocessing and NMS.

  • July 23, 2020: v2.0 release: improved model definition, training and mAP.
  • June 22, 2020: PANet updates: new heads, reduced parameters, improved speed and mAP 364fcfd.
  • June 19, 2020: FP16 as new default for smaller checkpoints and faster inference d4c6674.
  • June 9, 2020: CSP updates: improved speed, size, and accuracy (credit to @WongKinYiu for CSP).
  • May 27, 2020: Public release. YOLOv5 models are SOTA among all known YOLO implementations.
  • April 1, 2020: Start development of future compound-scaled YOLOv3/YOLOv4-based PyTorch models.

Pretrained Checkpoints

Model APval APtest AP50 SpeedGPU FPSGPU params FLOPS
YOLOv5s 36.1 36.1 55.3 2.1ms 476 7.5M 13.2B
YOLOv5m 43.5 43.5 62.5 3.0ms 333 21.8M 39.4B
YOLOv5l 47.0 47.1 65.6 3.9ms 256 47.8M 88.1B
YOLOv5x 49.0 49.0 67.4 6.1ms 164 89.0M 166.4B
YOLOv3-SPP 45.6 45.5 65.2 4.5ms 222 63.0M 118.0B

** APtest denotes COCO test-dev2017 server results, all other AP results in the table denote val2017 accuracy.
** All AP numbers are for single-model single-scale without ensemble or test-time augmentation. Reproduce by python test.py --data coco.yaml --img 672 --conf 0.001
** SpeedGPU measures end-to-end time per image averaged over 5000 COCO val2017 images using a GCP n1-standard-16 instance with one V100 GPU, and includes image preprocessing, PyTorch FP16 image inference at --batch-size 32 --img-size 640, postprocessing and NMS. Average NMS time included in this chart is 1-2ms/img. Reproduce by python test.py --data coco.yaml --img 640 --conf 0.1
** All checkpoints are trained to 300 epochs with default settings and hyperparameters (no autoaugmentation).

v1.0

3 years ago

YOLOv5 1.0 Release Notes

  • June 22, 2020: PANet updates: increased layers, reduced parameters, faster inference and improved mAP 364fcfd.
  • June 19, 2020: FP16 as new default for smaller checkpoints and faster inference d4c6674.
  • June 9, 2020: CSP updates: improved speed, size, and accuracy. Credit to @WongKinYiu for excellent CSP work.
  • May 27, 2020: Public release of repo. YOLOv5 models are SOTA among all known YOLO implementations.
  • April 1, 2020: Start development of future YOLOv3/YOLOv4-based PyTorch models in a range of compound-scaled sizes.
** GPU Speed measures end-to-end time per image averaged over 5000 COCO val2017 images using a V100 GPU with batch size 8, and includes image preprocessing, PyTorch FP16 inference, postprocessing and NMS.

Pretrained Checkpoints

Model APval APtest AP50 SpeedGPU FPSGPU params FLOPS
YOLOv5s 36.6 36.6 55.8 2.1ms 476 7.5M 13.2B
YOLOv5m 43.4 43.4 62.4 3.0ms 333 21.8M 39.4B
YOLOv5l 46.6 46.7 65.4 3.9ms 256 47.8M 88.1B
YOLOv5x 48.4 48.4 66.9 6.1ms 164 89.0M 166.4B
YOLOv3-SPP 45.6 45.5 65.2 4.5ms 222 63.0M 118.0B

** APtest denotes COCO test-dev2017 server results, all other AP results in the table denote val2017 accuracy.
** All AP numbers are for single-model single-scale without ensemble or test-time augmentation. Reproduce by python test.py --img 736 --conf 0.001
** SpeedGPU measures end-to-end time per image averaged over 5000 COCO val2017 images using a GCP n1-standard-16 instance with one V100 GPU, and includes image preprocessing, PyTorch FP16 image inference at --batch-size 32 --img-size 640, postprocessing and NMS. Average NMS time included in this chart is 1-2ms/img. Reproduce by python test.py --img 640 --conf 0.1
** All checkpoints are trained to 300 epochs with default settings and hyperparameters (no autoaugmentation).