Day48 - Backbone中C3/C2f后添加一层SEAttention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SE,在FLIR上训练200epochs
Yolov5n-se
Parameters: 2.666M
Class Precision Recall mAP50 mAP50-95 all 0.587 0.438 0.483 0.29 person 0.831 0.608 0.741 0.415 bike 0.495 0.447 0.415 0.258 car 0.836 0.744 0.83 0.585 motor 0.749 0.582 0.653 0.35 bus 0.796 0.419 0.578 0.396 truck 0.222 0.299 0.146 0.0889 light 0.717 0.392 0.481 0.219 other vehicle 0.0461 0.0159 0.0212 0.00766 Yolov5s-se
Parameters: 9.197M
Class Precision Recall mAP50 mAP50-95 all 0.65 0.497 0.537 0.331 person 0.858 0.654 0.784 0.463 bike 0.567 0.512 0.517 0.318 car 0.874 0.774 0.865 0.624 motor 0.86 0.582 0.672 0.362 bus 0.784 0.527 0.654 0.466 truck 0.354 0.413 0.209 0.138 light 0.79 0.464 0.566 0.269 other vehicle 0.11 0.0476 0.0278 0.00848 Yolov5m-se
Parameters: 25.209M
Class Precision Recall mAP50 mAP50-95 all 0.67 0.521 0.574 0.364 person 0.893 0.666 0.811 0.497 bike 0.644 0.565 0.55 0.355 car 0.9 0.781 0.882 0.65 motor 0.757 0.655 0.711 0.394 bus 0.826 0.478 0.664 0.485 truck 0.371 0.478 0.291 0.194 light 0.825 0.499 0.623 0.304 other vehicle 0.142 0.0476 0.0617 0.0322 Yolov8n-se
Parameters: 3.168M
Class Precision Recall mAP50 mAP50-95 all 0.619 0.449 0.488 0.292 person 0.83 0.61 0.744 0.42 bike 0.479 0.435 0.399 0.24 car 0.848 0.752 0.843 0.598 motor 0.83 0.582 0.632 0.319 bus 0.835 0.441 0.62 0.432 truck 0.256 0.321 0.169 0.101 light 0.715 0.389 0.467 0.211 other vehicle 0.158 0.0635 0.0297 0.0154 Yolov8s-se
Parameters: 11.210M
Class Precision Recall mAP50 mAP50-95 all 0.624 0.505 0.544 0.338 person 0.846 0.675 0.792 0.471 bike 0.543 0.541 0.514 0.328 car 0.87 0.785 0.87 0.635 motor 0.736 0.6 0.678 0.385 bus 0.786 0.513 0.667 0.465 truck 0.268 0.348 0.202 0.136 light 0.767 0.479 0.56 0.26 other vehicle 0.175 0.0952 0.0683 0.0237
本文由作者按照 CC BY 4.0 进行授权