search

mingyuan-zhang / MotionDiffuse

MotionDiffuse: Text-Driven Human Motion Generation with Diffusion Model
https://mingyuan-zhang.github.io/projects/MotionDiffuse.html
Other
835 stars 74 forks source link

About training time and hardware #11

Closed SMY19999 closed 1 year ago

SMY19999 commented 1 year ago

Brilliant work as is shown in this project! However, I'd be appreciate that if you could show your GPU type and total training time.

mingyuan-zhang commented 1 year ago

Hi, officially, we use 8 NVIDIA Tesla V100 SXM2(32 GB) to train our models. As for KIT-ML, the whole training costs 10 hours. HumanML3D requires 2 days for training. We are now trying to decrease the batchsize and retrain the number of iterations. We will update these results then.

SMY19999 commented 1 year ago

Thanks a lot ! It is really amazing!