search

ChenFengYe / motion-latent-diffusion

[CVPR 2023] Executing your Commands via Motion Diffusion in Latent Space, a fast and high-quality motion diffusion model
https://chenxin.tech/mld/
MIT License
586 stars 55 forks source link

Question about training on humanml3d dataset #20

Closed HoBeom closed 1 year ago

HoBeom commented 1 year ago

Thank you for sharing good research results and code.

Attempting to learn in humanml3d. However, learning is interrupted during validation because nan values in motion data cannot be handled.

How can I exclude nan values from the motion data?

Thank you.

https://github.com/ChenFengYe/motion-latent-diffusion/blob/719c219bd059e14f84f2571ab6975855ede0d819/mld/data/humanml/data/dataset.py#L437-L438

ChenFengYe commented 1 year ago

Hi, the nan might happen when you process your own humanml3d data. You can email the author of humanml3d for the correct dataset, or just remove the nan file. It could only happen in several npy or json files. It will not influence the whole experiment.

HoBeom commented 1 year ago

Thank you for your quick reply. As you expected, there was a problem with the dataset. I solved it and left a humanml3d issue. https://github.com/EricGuo5513/HumanML3D/issues/44 Thank you.