Why Yolov10 performance on custom dataset worse than yolov7. What hyperparameters can be adjusted to improve performance?

[mazatov]

I'm training on my custom object detection dataset. I'm detecting a ball in sports images. So I have one object that is quite small!

I trained on the same model Yolov10-B and default Yolov7. The performance of Yolov10 was quite a bit worse, which surprised me. My guess is some hyperparameters might need adjusting. One thing I noticed was that Yolov10 exited after 189 epochs because no improvement was observed in 100 epochs while Yolov7 trained for all 300 epochs.

What hyperparameters can be adjusted to improve performance?

[HumamHelfawi]

I have the same problem. YoloV8 perform both better in term of accuracy and performance than YoloV10. I am training and testing using Ultralytics