self learning on yolov5 experinmented on coco128
docker build -f Dockerfile .
docker run --gpus all --shm-size=8g -v /to_yolo_dir:/workspace -it created_image_name
cd /workspace/yolo_something_name
python train.py --img 640 --batch 16 --epochs 3 --data coco128.yaml --weights yolov5s.pt --cache
will get error message and go datasets dir copy ../datasets/coco128/image/train2017 to ../datasets/coco128/image/unlabeled
python train.py --img 640 --batch 16 --epochs 300 --data coco128.yaml --weights yolov5s.pt --cache
supervised learning | self learning | |
---|---|---|
m@p0.5 m@p0.5-0.95 | m@p0.5 m@p0.5-0.95 | |
yolov5s | 0.967 0.799 | 0.968 0.824 |
yolov5m | 0.975 0.876 | 0.977 0.878 |
yolov5l | 0.977 0.878 | 0.979 0.914 |
yolov5x | 0.979 0.901 | 0.978 0.918 |