Question about t2m_mean.py and t2m_std.npy

GuyTevet / motion-diffusion-model

The official PyTorch implementation of the paper "Human Motion Diffusion Model"

MIT License

3.13k stars 338 forks source link

Question about t2m_mean.py and t2m_std.npy #197

Open yw0208 opened 6 months ago

yw0208 commented 6 months ago

Thanks for your great work! I noticed that you use different Mean and Std when training and evaluation. It seems that t2m_mean.py and t2m_std.npy are the mean and std of the output of MDM. I am confused that how did you get them? Do they have a significant impact on the final results? If I make improvements on MDM, does it mean that I need to obtain the t2m_mean.py and t2m_std.npy of my own model? Looking forward to your reply, thank you again.

GuyTevet commented 5 months ago

Hi @yw0208 , thanks! Indeed the evaluators were trained with a slightly different normalization (See https://github.com/EricGuo5513/text-to-motion). I don't think it makes a dramatic difference in performance but please double check me.