Day50 - Backbone中C3/C2f后添加一层ECA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ECA,在FLIR上训练200epochs
Yolov5n-eca
Parameters: 2.655M
Class Precision Recall mAP50 mAP50-95 all 0.556 0.449 0.474 0.283 person 0.793 0.642 0.733 0.405 bike 0.456 0.418 0.407 0.256 car 0.825 0.748 0.83 0.581 motor 0.67 0.564 0.576 0.291 bus 0.75 0.525 0.608 0.417 truck 0.237 0.283 0.145 0.0893 light 0.714 0.41 0.473 0.216 other vehicle 0 0 0.0191 0.00739 Yolov5s-eca
Parameters: 9.153M
Class Precision Recall mAP50 mAP50-95 all 0.602 0.531 0.547 0.335 person 0.808 0.692 0.777 0.455 bike 0.501 0.547 0.513 0.322 car 0.826 0.796 0.858 0.618 motor 0.74 0.655 0.7 0.358 bus 0.74 0.54 0.657 0.457 truck 0.333 0.457 0.269 0.191 light 0.731 0.483 0.557 0.26 other vehicle 0.139 0.0794 0.0467 0.018 Yolov5m-eca
Parameters: 25.111M
Class Precision Recall mAP50 mAP50-95 all 0.621 0.557 0.574 0.364 person 0.808 0.737 0.811 0.5 bike 0.597 0.541 0.526 0.34 car 0.831 0.835 0.884 0.654 motor 0.642 0.618 0.732 0.407 bus 0.795 0.564 0.687 0.505 truck 0.306 0.457 0.253 0.169 light 0.768 0.548 0.621 0.305 other vehicle 0.224 0.159 0.0774 0.0344 Yolov8n-eca
Parameters: 3.157M
Class Precision Recall mAP50 mAP50-95 all 0.593 0.457 0.494 0.297 person 0.827 0.621 0.746 0.422 bike 0.491 0.447 0.443 0.268 car 0.85 0.749 0.838 0.592 motor 0.62 0.582 0.607 0.307 bus 0.852 0.508 0.633 0.448 truck 0.208 0.263 0.141 0.0937 light 0.745 0.418 0.515 0.23 other vehicle 0.152 0.0635 0.0322 0.0128 Yolov8s-eca
Parameters: 11.167M
Class Precision Recall mAP50 mAP50-95 all 0.661 0.508 0.56 0.342 person 0.846 0.685 0.79 0.471 bike 0.615 0.553 0.535 0.332 car 0.853 0.801 0.869 0.634 motor 0.721 0.564 0.697 0.374 bus 0.774 0.537 0.689 0.479 truck 0.388 0.348 0.228 0.13 light 0.775 0.482 0.574 0.273 other vehicle 0.318 0.0952 0.0977 0.0435
本文由作者按照 CC BY 4.0 进行授权