How to Training

Training custom YOLO model

1. Collecting image data. Since there is almost no sign language image data open on the web for download and use, we took and collected images ourselves.

2. Labeling the images. In order to train images on a model, images must be labeled using a labeling tool. 'labelImg' was used as a labeling tool. When using labelImg, the mode should be set to YOLO so that the result is stored in a txt file.

• Installation and using labelImg : https://github.com/tzutalin/labelImg

3. The following files are required to train custom data.

obj.data     
obj.cfg    
obj.names    
train.txt   
valid.txt    

4. After downloading Yolov4's pretrained model(yolov4.conv.137) provided by darknet, training begins.

5. train YOLOv4

./darknet detector train custom/obj.data custom/obj.cfg yolov4.conv.137 -map

6. Accuracy

Testing

After downloading darknet.py and sonmari_video.py, run sonmari_video.py. Weights file obtained as a result of training is used.

sonmari_video.py