Publication | Open Access
ultralytics/yolov5: v6.0 - YOLOv5n 'Nano' models, Roboflow integration, TensorFlow export, OpenCV DNN support
207
Citations
0
References
2021
Year
This release incorporates many new features and bug fixes (<strong>465 PRs</strong> from <strong>73 contributors</strong>) since our last release v5.0 in April, brings architecture tweaks, and also introduces new P5 and P6 'Nano' models: <strong>YOLOv5n</strong> and <strong>YOLOv5n6</strong>. Nano models maintain the YOLOv5s depth multiple of 0.33 but reduce the YOLOv5s width multiple from 0.50 to 0.25, resulting in ~75% fewer parameters, from 7.5M to 1.9M, ideal for mobile and CPU solutions. Example usage: <pre><code class="lang-bash">python detect.py --weights yolov5n.pt --img 640 # Nano P5 model trained at --img 640 (28.4 mAP@0.5:0.95) python detect.py --weights yolov5n6.pt --img 1280 # Nano P6 model trained at --img 1280 (34.0 mAP0.5:0.95) </code></pre> Important Updates <strong>Roboflow Integration ⭐ NEW</strong>: Train YOLOv5 models directly on any Roboflow dataset with our new integration! (https://github.com/ultralytics/yolov5/issues/4975 by @Jacobsolawetz) <strong>YOLOv5n 'Nano' models ⭐ NEW</strong>: New smaller YOLOv5n (1.9M params) model below YOLOv5s (7.5M params), exports to 2.1 MB INT8 size, ideal for ultralight mobile solutions. (https://github.com/ultralytics/yolov5/discussions/5027 by @glenn-jocher) <strong>TensorFlow and Keras Export</strong>: TensorFlow, Keras, TFLite, TF.js model export now fully integrated using <code>python export.py --include saved_model pb tflite tfjs</code> (https://github.com/ultralytics/yolov5/pull/1127 by @zldrobit) <strong>OpenCV DNN</strong>: YOLOv5 ONNX models are now compatible with both OpenCV DNN and ONNX Runtime (https://github.com/ultralytics/yolov5/pull/4833 by @SamFC10). <strong>Model Architecture:</strong> Updated backbones are slightly smaller, faster and more accurate. Replacement of <code>Focus()</code> with an equivalent <code>Conv(k=6, s=2, p=2)</code> layer (https://github.com/ultralytics/yolov5/issues/4825 by @thomasbi1) for improved exportability New <code>SPPF()</code> replacement for <code>SPP()</code> layer for reduced ops (https://github.com/ultralytics/yolov5/pull/4420 by @glenn-jocher) Reduction in P3 backbone layer <code>C3()</code> repeats from 9 to 6 for improved speeds Reorder places <code>SPPF()</code> at end of backbone Reintroduction of shortcut in the last <code>C3()</code> backbone layer Updated hyperparameters with increased mixup and copy-paste augmentation New Results <p align="left"><img width="800" src="https://user-images.githubusercontent.com/26833433/136901921-abcfcd9d-f978-4942-9b97-0e3f202907df.png"></p> <details> <summary>YOLOv5-P5 640 Figure (click to expand)</summary> <p align="left"><img width="800" src="https://user-images.githubusercontent.com/26833433/136763877-b174052b-c12f-48d2-8bc4-545e3853398e.png"></p> </details> <details> <summary>Figure Notes (click to expand)</summary> * **COCO AP val** denotes mAP@0.5:0.95 metric measured on the 5000-image [COCO val2017](http://cocodataset.org) dataset over various inference sizes from 256 to 1536. * **GPU Speed** measures average inference time per image on [COCO val2017](http://cocodataset.org) dataset using a [AWS p3.2xlarge](https://aws.amazon.com/ec2/instance-types/p3/) V100 instance at batch-size 32. * **EfficientDet** data from [google/automl](https://github.com/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` </details>mAP improves from +0.3% to +1.1% across all models, and ~5% FLOPs reduction produces slight speed improvements and a reduced CUDA memory footprint. Example YOLOv5l before and after metrics: YOLOv5l<br><sup>Large size<br><sup>(pixels) mAP<sup>val<br>0.5:0.95 mAP<sup>val<br>0.5 Speed<br><sup>CPU b1<br>(ms) Speed<br><sup>V100 b1<br>(ms) Speed<br><sup>V100 b32<br>(ms) params<br><sup>(M) FLOPs<br><sup>@640 (B) v5.0 (previous) 640 48.2 66.9 457.9 11.6 2.8 47.0 115.4 v6.0 (this release) 640 <strong>48.8</strong> <strong>67.2</strong> <strong>424.5</strong> <strong>10.9</strong> <strong>2.7</strong> <strong>46.5</strong> <strong>109.1</strong> Pretrained Checkpoints Model size<br><sup>(pixels) mAP<sup>val<br>0.5:0.95 mAP<sup>val<br>0.5 Speed<br><sup>CPU b1<br>(ms) Speed<br><sup>V100 b1<br>(ms) Speed<br><sup>V100 b32<br>(ms) params<br><sup>(M) FLOPs<br><sup>@640 (B) YOLOv5n 640 28.4 46.0 <strong>45</strong> <strong>6.3</strong> <strong>0.6</strong> <strong>1.9</strong> <strong>4.5</strong> YOLOv5s 640 37.2 56.0 98 6.4 0.9 7.2 16.5 YOLOv5m 640 45.2 63.9 224 8.2 1.7 21.2 49.0 YOLOv5l 640 48.8 67.2 430 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 34.0 50.7 153 8.1 2.1 3.2 4.6 YOLOv5s6 1280 44.5 63.0 385 8.2 3.6 16.8 12.6 YOLOv5m6 1280 51.0 69.0 887 11.1 6.8 35.7 50.0 YOLOv5l6 1280 53.6 71.6 1784 15.8 10.5 76.8 111.4 YOLOv5x6<br>+ TTA 1280<br>1536 54.7<br><strong>55.4</strong> <strong>72.4</strong><br>72.3 3136<br>- 26.2<br>- 19.4<br>- 140.7<br>- 209.8<br>- <details> <summary>Table Notes (click to expand)</summary> * All checkpoints are trained to 300 epochs with default settings. Nano models use [hyp.scratch-low.yaml](https://github.com/ultralytics/yolov5/blob/master/data/hyps/hyp.scratch-low.yaml) hyperparameters, all others use [hyp.scratch-high.yaml](https://github.com/ultralytics/yolov5/blob/master/data/hyps/hyp.scratch-high.yaml). * **mAP<sup>val</sup>** values are for single-model single-scale on [COCO val2017](http://cocodataset.org) dataset.<br>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](https://aws.amazon.com/ec2/instance-types/p3/) instance. NMS times (~1 ms/img) not included.<br>Reproduce by `python val.py --data coco.yaml --img 640 --conf 0.25 --iou 0.45` * **TTA** [Test Time Augmentation](https://github.com/ultralytics/yolov5/issues/303) includes reflection and scale augmentations.<br>Reproduce by `python val.py --data coco.yaml --img 1536 --iou 0.7 --augment` </details>Changelog Changes between <strong>previous release and this release</strong>: https://github.com/ultralytics/yolov5/compare/v5.0...v6.0 Changes <strong>since this release</strong>: https://github.com/ultralytics/yolov5/compare/v6.0...HEAD <details> <summary>New Features and Bug Fixes (465)</summary> * YOLOv5 v5.0 Release patch 1 by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2764 * Flask REST API Example by @robmarkcole in https://github.com/ultralytics/yolov5/pull/2732 * ONNX Simplifier by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2815 * YouTube Bug Fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2818 * PyTorch Hub cv2 .save() .show() bug fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2831 * Create FUNDING.yml by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2832 * Update FUNDING.yml by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2833 * Fix ONNX dynamic axes export support with onnx simplifier, make onnx simplifier optional by @timstokman in https://github.com/ultralytics/yolov5/pull/2856 * Update increment_path() to handle file paths by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2867 * Detection cropping+saving feature addition for detect.py and PyTorch Hub by @Ab-Abdurrahman in https://github.com/ultralytics/yolov5/pull/2827 * Implement yaml.safe_load() by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2876 * Cleanup load_image() by @JoshSong in https://github.com/ultralytics/yolov5/pull/2871 * bug fix: switched rows and cols for correct detections in confusion matrix by @MichHeilig in https://github.com/ultralytics/yolov5/pull/2883 * VisDrone2019-DET Dataset Auto-Download by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2882 * Uppercase model filenames enabled by @r-blmnr in https://github.com/ultralytics/yolov5/pull/2890 * ACON activation function by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2893 * Explicit opt function arguments by @fcakyon in https://github.com/ultralytics/yolov5/pull/2817 * Update yolo.py by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2899 * Update google_utils.py by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2900 * Add detect.py --hide-conf --hide-labels --line-thickness options by @Ashafix in https://github.com/ultralytics/yolov5/pull/2658 * Default optimize_for_mobile() on TorchScript models by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2908 * Update export.py onnx -> ct print bug fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2909 * Update export.py for 2 dry runs by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2910 * Add file_size() function by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2911 * Update download() for tar.gz files by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2919 * Update visdrone.yaml bug fix by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2921 * changed default value of hide label argument to False by @albinxavi in https://github.com/ultralytics/yolov5/pull/2923 * Change default value of hide-conf argument to false by @albinxavi in https://github.com/ultralytics/yolov5/pull/2925 * test.py native --single-cls by @glenn-jocher in https://github.com/ultralytics/yolov5/pull/2928 * Add verbose option to pytorch hub models by @NanoCode012 in https://github.com/ultralytics/yolov5/pull/2926 * ACON Activation batch-size 1 bug patch by @glenn-joche