YOLOv7-segmentation_holecover

Detail informaion can be browsed on my Medium.

1. Converts the format of VGG Image Annotator (VIA) to the YOLOv7 segmentation format.

Classes:manhole, handhole
code of format coversion: link

2. Modify data.yaml

Modify a yaml file for the classes of the custom dataset in yolov7-mask/data/
give images and labels path and classes.

3. Training

  python segment/train.py --data ./data/data.yaml --batch 16 --weights ./runs/train-seg/yolov7-seg/weights/last.pt  --cfg yolov7-seg.yaml --epochs 100 --name yolov7-seg --img 640 --hyp hyp.scratch-high.yaml

4. Evaluation

  python segment/predict.py --weights ./runs/train-seg/yolov7-seg2/weights/best.pt --source /content/drive/MyDrive/final_hole/images/val/170727_020736227_Camera_4.jpg --iou-thres 0.5 --conf-thres 0.6

5. Visulization

  python segment/val.py --weights ./runs/train-seg/yolov7-seg2/weights/best.pt --data ./data/data.yaml --iou-thres 0.5 --conf-thres 0.6

6. Result

Manhole

Handhole