I am currently using YOLOv7 with the COCO dataset weights to detect two classes - ‘Person’ and ‘Sports Ball’ from football match footage. While it performs well in detecting people, the accuracy for sports ball detection is relatively poor. To address this, I’ve acquired the SoccernetV3 dataset, containing approximately 10,000 sports ball images.
Could you please guide me on how to fine-tune the YOLOv7 model using the SoccernetV3 dataset while maintaining the current level of accuracy in detecting humans? My goal is to enhance sports ball detection without compromising human detection accuracy. Thank you!
