orgilj/semi-yolov5

semi-supervised yolov5

self-yolov5

self learning on yolov5 experinmented on coco128

build docker image

docker build -f Dockerfile .

docker run --gpus all --shm-size=8g -v /to_yolo_dir:/workspace -it created_image_name

in container change directory /workspace/yolo directory

cd /workspace/yolo_something_name

download coco128 dataset

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

also in our experiment we copied coco validation dataset to unlabeled directory

after copying dataset run train.py again

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

if there is shape size error appears just ignore them (actualy requires label error)