Day38 - Neck中C3/C2f后添加一层CBAM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CBAM,在FLIR上训练200epochs
Yolov5n-cbam
Parameters: 2.758M
Class Precision Recall mAP50 mAP50-95 all 0.571 0.472 0.488 0.288 person 0.792 0.642 0.741 0.412 bike 0.471 0.441 0.412 0.256 car 0.8 0.763 0.832 0.585 motor 0.676 0.618 0.653 0.313 bus 0.783 0.484 0.61 0.402 truck 0.234 0.37 0.174 0.116 light 0.664 0.393 0.455 0.207 other vehicle 0.151 0.065 0.0295 0.0106 Yolov5s-cbam
Parameters: 9.564M
Class Precision Recall mAP50 mAP50-95 all 0.651 0.489 0.541 0.333 person 0.874 0.634 0.783 0.463 bike 0.562 0.494 0.475 0.29 car 0.886 0.766 0.864 0.624 motor 0.827 0.607 0.715 0.359 bus 0.781 0.48 0.636 0.461 truck 0.315 0.37 0.232 0.164 light 0.796 0.478 0.573 0.275 other vehicle 0.168 0.0794 0.0502 0.0269 Yolov5m-cbam
Parameters: 26.035M
Class Precision Recall mAP50 mAP50-95 all 0.696 0.5 0.569 0.362 person 0.899 0.647 0.806 0.5 bike 0.699 0.535 0.553 0.362 car 0.906 0.78 0.882 0.652 motor 0.828 0.6 0.694 0.401 bus 0.852 0.48 0.647 0.47 truck 0.361 0.413 0.297 0.195 light 0.844 0.483 0.622 0.298 other vehicle 0.179 0.0635 0.0516 0.0218 Yolov8n-cbam
Parameters: 3.260M
Class Precision Recall mAP50 mAP50-95 all 0.584 0.444 0.486 0.294 person 0.818 0.625 0.748 0.427 bike 0.434 0.459 0.421 0.259 car 0.845 0.756 0.842 0.596 motor 0.759 0.545 0.616 0.313 bus 0.783 0.419 0.59 0.421 truck 0.207 0.326 0.142 0.089 light 0.771 0.403 0.513 0.237 other vehicle 0.0543 0.0159 0.0199 0.00801 Yolov8s-cbam
Parameters: 11.577M
Class Precision Recall mAP50 mAP50-95 all 0.646 0.513 0.558 0.346 person 0.852 0.665 0.786 0.472 bike 0.577 0.553 0.543 0.353 car 0.868 0.79 0.87 0.631 motor 0.787 0.618 0.71 0.38 bus 0.8 0.537 0.668 0.476 truck 0.344 0.413 0.259 0.162 light 0.766 0.462 0.55 0.259 other vehicle 0.176 0.0677 0.0773 0.0293
本文由作者按照 CC BY 4.0 进行授权