Day31 - P2改进在FLIR上的表现
yolov5n/s/m和yolov8n/s仅添加p2检测头,在FLIR上训练200epochs
Yolov5n-p2
Parameters: 2.838M
Class Precision Recall mAP50 mAP50-95 all 0.613 0.467 0.511 0.306 person 0.838 0.662 0.794 0.459 bike 0.525 0.441 0.435 0.256 car 0.857 0.773 0.865 0.618 motor 0.801 0.545 0.604 0.303 bus 0.748 0.469 0.629 0.43 truck 0.237 0.304 0.14 0.0806 light 0.758 0.492 0.578 0.279 other vehicle 0.135 0.0476 0.0401 0.0238 Yolov5s-p2
Parameters: 8.816M
Class Precision Recall mAP50 mAP50-95 all 0.637 0.525 0.572 0.358 person 0.873 0.697 0.836 0.507 bike 0.612 0.506 0.506 0.32 car 0.876 0.809 0.891 0.654 motor 0.624 0.604 0.658 0.348 bus 0.853 0.503 0.692 0.487 truck 0.261 0.457 0.262 0.183 light 0.823 0.544 0.659 0.329 other vehicle 0.174 0.0794 0.0713 0.0327 Yolov5m-p2
Parameters: 23.863M
Class Precision Recall mAP50 mAP50-95 all 0.683 0.539 0.596 0.384 person 0.901 0.694 0.852 0.53 bike 0.628 0.566 0.551 0.36 car 0.907 0.805 0.901 0.669 motor 0.765 0.651 0.713 0.406 bus 0.784 0.52 0.679 0.493 truck 0.334 0.413 0.299 0.218 light 0.836 0.566 0.696 0.35 other vehicle 0.306 0.0952 0.0773 0.0482 Yolov8n-p2
Parameters: 3.411M
Class Precision Recall mAP50 mAP50-95 all 0.592 0.475 0.528 0.32 person 0.809 0.693 0.805 0.471 bike 0.486 0.467 0.455 0.279 car 0.841 0.796 0.872 0.627 motor 0.753 0.545 0.67 0.331 bus 0.748 0.447 0.625 0.449 truck 0.19 0.326 0.156 0.101 light 0.732 0.479 0.573 0.275 other vehicle 0.179 0.0476 0.0672 0.0254 Yolov8s-p2
Parameters: 11.110M
Class Precision Recall mAP50 mAP50-95 all 0.658 0.517 0.574 0.357 person 0.873 0.7 0.838 0.512 bike 0.512 0.529 0.508 0.332 car 0.87 0.812 0.891 0.656 motor 0.849 0.582 0.702 0.369 bus 0.802 0.508 0.68 0.483 truck 0.334 0.348 0.235 0.142 light 0.783 0.532 0.626 0.311 other vehicle 0.24 0.127 0.109 0.0546
本文由作者按照 CC BY 4.0 进行授权