These visualization functions are called during training, if we are only predicting one joint (one channel), then the old functions will fail. The fix allows to run LEAP normally with predicting only one joint in the skeleton.
The squeeze function with no axis argument was removing the channel dimension (of size 1 for one joint).
talmo
commented
5 years ago
These visualization functions are called during training, if we are only predicting one joint (one channel), then the old functions will fail. The fix allows to run LEAP normally with predicting only one joint in the skeleton.
The squeeze function with no axis argument was removing the channel dimension (of size 1 for one joint).