Based on the way that for-loops handle the creation of the data in extract_shape.py (basically rewriting existing variables at each frame), I suspected that only the last mesh was taken into account.
I looked into the NPZ for the YBot, and indeed, it only has 10514 vertices in rest_vertices, which is exactly the number of vertices in the "Joints" mesh of the XBot (see attached figure, in the bottom-right corner).
Does the method only expect character with a single mesh ? Maybe that explains some poor results I've had recently when trying to run the inference on provided files...?
Based on the way that for-loops handle the creation of the data in
extract_shape.py
(basically rewriting existing variables at each frame), I suspected that only the last mesh was taken into account.I looked into the NPZ for the YBot, and indeed, it only has 10514 vertices in
rest_vertices
, which is exactly the number of vertices in the "Joints" mesh of the XBot (see attached figure, in the bottom-right corner).Does the method only expect character with a single mesh ? Maybe that explains some poor results I've had recently when trying to run the inference on provided files...?