Day43 - Backbone中C3/C2f后添加一层GAM在FLIR上的表现
yolov5n/s/m和yolov8n/s仅在backbone的C3/C2f模块后添加一层GAM,在FLIR上训练200epochs
Yolov5n-gam
Parameters: 3.234M
Class Precision Recall mAP50 mAP50-95 all 0.599 0.473 0.492 0.296 person 0.805 0.624 0.738 0.413 bike 0.433 0.494 0.427 0.26 car 0.803 0.766 0.838 0.592 motor 0.804 0.545 0.62 0.316 bus 0.717 0.492 0.607 0.443 truck 0.29 0.37 0.176 0.111 light 0.703 0.416 0.482 0.219 other vehicle 0.236 0.0794 0.0463 0.0154 Yolov5s-gam
Parameters: 11.465M
Class Precision Recall mAP50 mAP50-95 all 0.635 0.521 0.546 0.341 person 0.837 0.686 0.796 0.47 bike 0.593 0.512 0.527 0.343 car 0.861 0.795 0.871 0.635 motor 0.767 0.564 0.664 0.362 bus 0.763 0.497 0.65 0.466 truck 0.309 0.543 0.228 0.155 light 0.768 0.476 0.574 0.274 other vehicle 0.182 0.0952 0.0608 0.0263 Yolov5m-gam
Parameters: 30.308M
Class Precision Recall mAP50 mAP50-95 all 0.685 0.516 0.587 0.372 person 0.897 0.654 0.813 0.502 bike 0.63 0.529 0.563 0.361 car 0.905 0.778 0.884 0.655 motor 0.786 0.673 0.781 0.438 bus 0.828 0.48 0.662 0.481 truck 0.393 0.478 0.302 0.203 light 0.837 0.48 0.609 0.298 other vehicle 0.205 0.0574 0.0809 0.0348 Yolov8n-gam
Parameters: 3.736M
Class Precision Recall mAP50 mAP50-95 all 0.564 0.468 0.494 0.3 person 0.79 0.646 0.744 0.416 bike 0.481 0.465 0.445 0.268 car 0.808 0.773 0.842 0.596 motor 0.674 0.582 0.635 0.361 bus 0.722 0.478 0.6 0.423 truck 0.209 0.326 0.145 0.0917 light 0.675 0.413 0.47 0.211 other vehicle 0.15 0.0635 0.0694 0.033 Yolov8s-gam
Parameters: 13.478M
Class Precision Recall mAP50 mAP50-95 all 0.635 0.516 0.544 0.339 person 0.848 0.684 0.797 0.473 bike 0.575 0.553 0.515 0.32 car 0.851 0.8 0.869 0.632 motor 0.757 0.6 0.683 0.378 bus 0.771 0.525 0.642 0.462 truck 0.28 0.391 0.207 0.134 light 0.752 0.496 0.572 0.273 other vehicle 0.244 0.0794 0.0661 0.0381
本文由作者按照 CC BY 4.0 进行授权