Open giacomovec opened 2 months ago
Hi,
For differences in scale, you would need to resize the keypoints to a common scale before modeling them.
Can you elaborate on what you mean by the rotation not matching?
I think the rotation is already handled by your code. For the difference in scale, I read that you were planning to implement a step to normalize them. Would that work? Do you know when will that be ready? Thank you very much for your answer
The method we are implementing is designed for more complicated normalization like differences in body shape. In your case, uniform scaling should work. You could use average statistics like nose-tail distance to determine the scale of each video and then uniformly scale the keypoints so they all share a common scale.
Hello, I have many videos with tracked points. My problem is that every video is from a different setup, so the rotation and scale of the keypoints doesn't match. How would I use your package with videos taken from different setups and cameras?