Performance Degradation in HumanML3D

Hello,
▪️I followed the instructions on GitHub step by step to prepare the [HumanML3D dataset ](https://github.com/GuyTevet/motion-diffusion-model?tab=readme-ov-file#b-full-data-text--motion-capture) and conducted an evaluation.


▪️I only performed the evaluation using the humanml-encoder-512 pretrained model, yet I observed a **performance degradation.**


▪️It seems like there is an issue with the dataset or the process, but I am unable to determine the cause, so I would like to ask for help.

For reference, I followed all the instructions for HUMANML3D, and the number of files in motion-diffusion-model/dataset/HumanML3D/new_joint_vecs was 29,228.

Best regards

Start with the README's HumanML3D preparation instructions and reproduce the evaluation using the humanml-encoder-512 pretrained model. Check the dataset/HumanML3D/new_joint_vecs contents, including the reported 29,228 files, against the evaluation inputs and reported results; done means identifying whether the degradation comes from the dataset or preparation process and documenting the cause.

Written by the indexing model from the issue text.