Loss nan for customized data

EricGuo5513 / momask-codes

Official implementation of "MoMask: Generative Masked Modeling of 3D Human Motions (CVPR2024)"

https://ericguo5513.github.io/momask/

MIT License

690 stars 56 forks source link

Closed hdjsjyl closed 5 months ago

hdjsjyl commented 5 months ago

Hi Author, While I train vaq on my customized data, the loss is nan. Any suggestion would be appreciated, thanks

Murrol commented 5 months ago

You may want to inspect your data to identify any occurrences of NaN or Inf values.

hdjsjyl commented 5 months ago

thanks

AlbertBJ commented 4 months ago

thanks

I'm beginer for this field; I want to know how to convert VIBE output to HumanML3D format using my own recorded video. thanks