When training YOLOv5 model with original dataloaders.py, you may notice images which have incomplete Ground-Truth-Boxes are used.
Here is modified dataloaders.py which keeps Ground-Truth-Boxes as you defined even with Mosaic augmentation.
- yv5_60 for YOLOv5 version 6.0
- yv5_70 for YOLOv5 version 7.0
- yv8 for YOLOv8