YOLOv5 Series Multi-backbone(TPH-YOLOv5, Ghostnet), Pruning and quantization Compression Tool Box
Update News
2021.10.30 复现TPH-YOLOv5
2021.10.31 完成替换backbone为Ghostnet
Requirements
环境安装
pip install -r requirements.txt
Evaluation metric
Visdrone DataSet (1-5 size is 608，6-8 size is 640)
| Model | mAP | [email protected] | Parameters(M) | GFLOPs | [email protected] |
|---|---|---|---|---|---|
| YOLOv5n | 13 | 26.2 | 1.78 | 4.2 | |
| YOLOv5s | 18.4 | 34 | 7.05 | 15.9 | |
| YOLOv5m | 21.6 | 37.8 | 20.91 | 48.2 | |
| YOLOv5l | 23.2 | 39.7 | 46.19 | 108.1 | |
| YOLOv5x | 24.3 | 40.8 | 86.28 | 204.4 | |
| YOLOv5xP2 | 30.00 | 49.29 | 90.96 | 314.2 | |
| YOLOv5xP2 CBAM | 30.13 | 49.40 | 91.31 | 315.1 | |
| YOLOv5xP2 CBAM TPH | 86.08 | 238.9 |
训练脚本实例：