search

theEricMa / OTAvatar

This is the official repository for OTAvatar: One-shot Talking Face Avatar with Controllable Tri-plane Rendering [CVPR2023].
306 stars 37 forks source link

About training consumption and inference speed #23

Closed silence-tang closed 10 months ago

silence-tang commented 10 months ago

Great work, it inspired me a lot!

May I ask the GPU memory needed to train the model? I don't have many GPUs and I'm afraid that the experiment can't be reproduced normally.

Besides, I'm also curious about the inference time (FPS).

Looking forward to hearing from you, thanks!

theEricMa commented 10 months ago

Thank you for your interest in our research on 3D talking faces. We recognize there are many concurrent efforts focused on creating generalizable and one-shot 3D talking face models. However, our approach stands out in its efficiency, requiring less GPU usage than other methods. Rather than training a model from scratch, we build on top of a pre-trained 3D GAN to achieve controllable facial motion. Specific details on the GPU requirements and inference times of our method can be found in our paper and demo video, respectively.