GreatV / labelme2yolo

Efficiently converts LabelMe's JSON format to the YOLOv5 dataset format.
MIT License
50 stars 14 forks source link

Background images are removed #71

Open janbrueggenjuergen opened 1 day ago

janbrueggenjuergen commented 1 day ago

Yolo models such as yolov9 and its implementation by WongKinYiu accept background images during training not containing any classes. This is usually solved by mixing images with blank labels or just the images into the dataset.

The current implementation sorts out images with no corresponding label file.

Maybe a switch to include images without labels and maybe one to generate empty label files would be a great enhancement.

Best regards

GreatV commented 1 day ago

fixed by