Day37 - Backbone中C3/C2f后添加一层CBAM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CBAM,在FLIR上训练200epochs
Yolov5n-cbam
Parameters: 2.742M
Class Precision Recall mAP50 mAP50-95 all 0.578 0.463 0.494 0.297 person 0.804 0.632 0.737 0.411 bike 0.529 0.429 0.433 0.261 car 0.822 0.763 0.837 0.589 motor 0.696 0.564 0.632 0.336 bus 0.769 0.508 0.634 0.427 truck 0.243 0.378 0.162 0.116 light 0.715 0.418 0.497 0.227 other vehicle 0.0483 0.0159 0.0183 0.00953 Yolov5s-cbam
Parameters: 9.502M
Class Precision Recall mAP50 mAP50-95 all 0.652 0.503 0.556 0.342 person 0.862 0.649 0.787 0.465 bike 0.566 0.512 0.518 0.327 car 0.878 0.779 0.868 0.629 motor 0.887 0.573 0.732 0.384 bus 0.809 0.52 0.667 0.465 truck 0.363 0.478 0.241 0.171 light 0.772 0.481 0.574 0.272 other vehicle 0.0789 0.0317 0.0633 0.0255 Yolov5m-cbam
Parameters: 25.896M
Class Precision Recall mAP50 mAP50-95 all 0.688 0.525 0.591 0.372 person 0.878 0.67 0.81 0.5 bike 0.653 0.518 0.548 0.356 car 0.897 0.79 0.882 0.653 motor 0.797 0.655 0.776 0.408 bus 0.838 0.514 0.687 0.5 truck 0.433 0.481 0.33 0.22 light 0.846 0.493 0.631 0.303 other vehicle 0.16 0.0794 0.0627 0.0346 Yolov8n-cbam
Parameters: 3.245M
Class Precision Recall mAP50 mAP50-95 all 0.597 0.462 0.499 0.303 person 0.808 0.626 0.743 0.421 bike 0.569 0.447 0.454 0.269 car 0.831 0.752 0.837 0.591 motor 0.844 0.564 0.642 0.354 bus 0.732 0.52 0.625 0.437 truck 0.267 0.37 0.182 0.114 light 0.694 0.399 0.468 0.221 other vehicle 0.0349 0.0159 0.039 0.0152 Yolov8s-cbam
Parameters: 19.245M
Class Precision Recall mAP50 mAP50-95 all 0.678 0.496 0.573 0.36 person 0.891 0.63 0.791 0.472 bike 0.607 0.488 0.544 0.329 car 0.907 0.751 0.865 0.628 motor 0.757 0.636 0.731 0.407 bus 0.816 0.522 0.673 0.496 truck 0.435 0.435 0.338 0.243 light 0.811 0.443 0.563 0.269 other vehicle 0.201 0.0635 0.0745 0.0395
本文由作者按照 CC BY 4.0 进行授权