Day49 - Backbone中C3/C2f后添加一层CA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CA,在FLIR上训练200epochs
Yolov5n-ca
Parameters: 2.667M
Class Precision Recall mAP50 mAP50-95 all 0.628 0.443 0.494 0.292 person 0.836 0.607 0.742 0.411 bike 0.537 0.418 0.41 0.245 car 0.853 0.732 0.833 0.587 motor 0.804 0.521 0.623 0.302 bus 0.801 0.475 0.619 0.417 truck 0.33 0.391 0.218 0.141 light 0.748 0.37 0.479 0.224 other vehicle 0.116 0.0317 0.0291 0.0122 Yolov5s-ca
Parameters: 9.191M
Class Precision Recall mAP50 mAP50-95 all 0.628 0.513 0.557 0.338 person 0.836 0.678 0.785 0.465 bike 0.587 0.512 0.521 0.321 car 0.859 0.788 0.865 0.627 motor 0.748 0.6 0.724 0.38 bus 0.746 0.508 0.655 0.439 truck 0.39 0.478 0.286 0.184 light 0.777 0.509 0.583 0.277 other vehicle 0.0765 0.0317 0.0361 0.0122 Yolov5m-ca
Parameters: 25.191M
Class Precision Recall mAP50 mAP50-95 all 0.694 0.527 0.575 0.363 person 0.884 0.659 0.808 0.49 bike 0.621 0.559 0.561 0.35 car 0.901 0.788 0.881 0.648 motor 0.811 0.582 0.674 0.388 bus 0.807 0.531 0.672 0.481 truck 0.398 0.457 0.288 0.198 light 0.842 0.502 0.616 0.298 other vehicle 0.292 0.143 0.0975 0.0472 Yolov8n-ca
Parameters: 3.170M
Class Precision Recall mAP50 mAP50-95 all 0.583 0.481 0.505 0.3 person 0.786 0.665 0.749 0.422 bike 0.471 0.441 0.435 0.263 car 0.811 0.772 0.841 0.595 motor 0.767 0.636 0.671 0.332 bus 0.777 0.488 0.615 0.427 truck 0.232 0.37 0.213 0.131 light 0.67 0.43 0.48 0.22 other vehicle 0.152 0.0483 0.0382 0.0116 Yolov8s-ca
Parameters: 11.204M
Class Precision Recall mAP50 mAP50-95 all 0.654 0.515 0.557 0.345 person 0.852 0.665 0.788 0.472 bike 0.642 0.518 0.528 0.32 car 0.869 0.791 0.87 0.631 motor 0.86 0.636 0.723 0.394 bus 0.731 0.531 0.671 0.472 truck 0.382 0.444 0.255 0.173 light 0.767 0.478 0.568 0.271 other vehicle 0.131 0.0601 0.0512 0.0233
本文由作者按照 CC BY 4.0 进行授权