Day42 - Neck中C3/C2f后添加一层GAM在FLIR上的表现
yolov5n/s/m和yolov8n/s仅在neck的C3/C2f模块后添加一层GAM,在FLIR上训练200epochs
Yolov5n-gam
Parameters: 3.336M
Class Precision Recall mAP50 mAP50-95 all 0.566 0.465 0.487 0.286 person 0.807 0.62 0.732 0.406 bike 0.462 0.435 0.428 0.245 car 0.816 0.76 0.833 0.581 motor 0.6 0.6 0.616 0.305 bus 0.781 0.477 0.621 0.422 truck 0.233 0.391 0.158 0.1 light 0.717 0.384 0.465 0.213 other vehicle 0.117 0.0524 0.0429 0.0163 Yolov5s-gam
Parameters: 11.873M
Class Precision Recall mAP50 mAP50-95 all 0.626 0.496 0.534 0.333 person 0.852 0.662 0.785 0.464 bike 0.628 0.518 0.506 0.319 car 0.862 0.785 0.863 0.625 motor 0.68 0.6 0.651 0.367 bus 0.798 0.486 0.642 0.452 truck 0.339 0.435 0.234 0.149 light 0.778 0.449 0.555 0.272 other vehicle 0.0731 0.0317 0.04 0.0137 Yolov5m-gam
Parameters: 31.226M
Class Precision Recall mAP50 mAP50-95 all 0.698 0.523 0.58 0.367 person 0.883 0.668 0.806 0.489 bike 0.648 0.531 0.545 0.356 car 0.906 0.782 0.878 0.644 motor 0.754 0.582 0.665 0.352 bus 0.872 0.52 0.674 0.487 truck 0.41 0.522 0.388 0.278 light 0.839 0.485 0.605 0.292 other vehicle 0.274 0.0952 0.0824 0.0403 Yolov8n-gam
Parameters: 3.838M
Class Precision Recall mAP50 mAP50-95 all 0.621 0.461 0.501 0.3 person 0.842 0.613 0.749 0.422 bike 0.463 0.488 0.464 0.276 car 0.852 0.753 0.838 0.595 motor 0.847 0.564 0.619 0.313 bus 0.805 0.439 0.601 0.412 truck 0.286 0.435 0.222 0.15 light 0.744 0.353 0.467 0.215 other vehicle 0.125 0.0476 0.0476 0.0185 Yolov8s-gam
Parameters: 13.886M
Class Precision Recall mAP50 mAP50-95 all 0.647 0.498 0.545 0.338 person 0.86 0.653 0.785 0.467 bike 0.515 0.512 0.492 0.323 car 0.883 0.779 0.866 0.63 motor 0.816 0.566 0.692 0.382 bus 0.798 0.536 0.663 0.461 truck 0.361 0.435 0.242 0.153 light 0.819 0.458 0.577 0.273 other vehicle 0.123 0.0476 0.0418 0.0186
本文由作者按照 CC BY 4.0 进行授权