NaN or Inf found in input tensor when runing trainIK.py

[fenaux]

After a random number of iterations I get
NaN or Inf found in input tensor

I made some tests and insert a test in "train_one_iteraton", when this occurs there NaN at the output of get_augmented_gt_seq_for_training

I also checked there are no NaN values in data

[aniongithub]

I'm seeing this issue too - has anyone figured out what's going on? @papagina any ideas? I'd love to reproduce these results for a project/paper I'm working on!

[WesleyXu224]

I meet the same situation!

[papagina]

This looks like a numerical issue but I didn't experience it. Could you share your torch version for me to try to reproduce the NaN situation?

[anklebreaker]

I too experienced this same problem on torch version 2.5.1.

I believe it's coming from line 453 in tools.py during rotation matrix to quaternion conversion as the radicand can be negative. A quick threshold fixes the issue, but it's probably better to have a more numerically robust conversion process.