- 验证集
命令:
python val.py --weights ./weights/yolov5l-xs-1.pt --img 1996 --data ./data/VisDrone.yaml --augment --save-txt --save-conf --task val --
batch-size 8 --verbose --name v5l-xs
结果:
val: Scanning '/disk2/lxs/tph-yolov5/datasets/VisDrone/VisDrone2019-DET-val/labels.cache' images and labels... 548 found, 0 missing, 0 empty, 0 corrupted: 100%|█| 548/548 [00:00<?, ?it
Class Images Labels P R mAP@.5 mAP@.5:.95: 100%|███████████████████████████████████████████████████████████| 69/69 [15:12<00:00, 13.23s/it]
all 548 38759 0.623 0.617 0.627 0.408
pedestrian 548 8844 0.678 0.723 0.745 0.408
people 548 5125 0.659 0.6 0.622 0.287
bicycle 548 1287 0.516 0.473 0.475 0.253
car 548 14064 0.78 0.908 0.917 0.697
van 548 1975 0.57 0.645 0.629 0.484
truck 548 750 0.545 0.64 0.603 0.441
tricycle 548 1045 0.596 0.489 0.513 0.327
awning-tricycle 548 532 0.472 0.256 0.297 0.208
bus 548 251 0.729 0.729 0.757 0.585
motor 548 4886 0.686 0.704 0.709 0.386
Speed: 1.0ms pre-process, 1544.0ms inference, 70.2ms NMS per image at shape (8, 3, 2016, 2016)
Results saved to runs/val/v5l-xs6
- 测试集
命令:
python val.py --weights ./weights/yolov5l-xs-1.pt --img 1996 --data ./data/VisDrone.yaml --augment --save-txt --save-conf --task test -
-batch-size 4 --verbose --name v5l-xs --device 0,1,2,3
结果:
test: Scanning '/disk2/lxs/tph-yolov5/datasets/VisDrone/VisDrone2019-DET-test-dev/labels' images and labels...1610 found, 0 missing, 0 empty, 0 corrupted: 100%|█| 1610/1610 [00:00<00:0
test: New cache created: /disk2/lxs/tph-yolov5/datasets/VisDrone/VisDrone2019-DET-test-dev/labels.cache
Class Images Labels P R mAP@.5 mAP@.5:.95: 100%|█████████████████████████████████████████████████████████| 403/403 [47:21<00:00, 7.05s/it]
all 1610 75102 0.565 0.541 0.525 0.322
pedestrian 1610 21006 0.588 0.544 0.553 0.247
people 1610 6376 0.577 0.348 0.366 0.141
bicycle 1610 1302 0.425 0.296 0.275 0.134
car 1610 28074 0.753 0.882 0.878 0.596
van 1610 5771 0.556 0.598 0.57 0.416
truck 1610 2659 0.532 0.685 0.641 0.449
tricycle 1610 530 0.385 0.513 0.377 0.224
awning-tricycle 1610 599 0.526 0.282 0.317 0.212
bus 1610 2940 0.738 0.674 0.73 0.551
motor 1610 5845 0.569 0.592 0.541 0.256
Speed: 1.1ms pre-process, 1658.1ms inference, 65.4ms NMS per image at shape (4, 3, 2016, 2016)