Bugs in evaluation code for processing language.

[Silverster98]

Hi,

I have noticed that there may be a bug in your modified evaluation code as follows.

motion-diffusion-model/data_loaders/humanml/motion_loaders/comp_v6_model_dataset.py

Line 214 in af061ca

'cap_len': len(tokens[bs_i]),

Because the tokens are all padded, if you use len(tokens[bs_i]) to obtain the cap_len, then all sentence lengths will be max_text_len=20 + 2. This will influence the language feature extraction for computing metrics.

And the following code is the original code in HumanML3D, which uses the right token length.

motion-diffusion-model/data_loaders/humanml/motion_loaders/comp_v6_model_dataset.py

Line 100 in af061ca

'cap_len': cap_lens[0].item(),

I think this bug may lead to a performance drop in MatchingScore, R-Precision, and so on.

[GuyTevet]

That's interesting, can you share the performance of the published model after your bug fix?

[GuyTevet]

Fixed. Thanks!