I was thinking about adding newer Pose Estimation networks and remove mediapipe. I tried an Vision Transformer, but the DLT doesn't work. The scaling of the points pxl_x, pxl_y is in the 10⁵ range. What else do you have to change besides the prediction network? Do you need to do some normalization of the points?
I was thinking about adding newer Pose Estimation networks and remove mediapipe. I tried an Vision Transformer, but the DLT doesn't work. The scaling of the points pxl_x, pxl_y is in the 10⁵ range. What else do you have to change besides the prediction network? Do you need to do some normalization of the points?