RTMO Bounding Box Accuracy

[orestis-z]

Hello, thanks a lot for the great work.

I evaluated the bounding box predictions of RTMO, and they seem relatively low. Kapao, on the other hand, performs well on both keypoints and bounding boxes. Surprisingly, the detection accuracy of RTMO is better on the small model compared to the large or medium one.

Is this a known issue?

[eeyrw]

It's said that small model does not prone to overfit and generalize better.