search

Fictionarry / ER-NeRF

[ICCV'23] Efficient Region-Aware Neural Radiance Fields for High-Fidelity Talking Portrait Synthesis
https://fictionarry.github.io/ER-NeRF/
MIT License
1.02k stars 133 forks source link

head shaking is noticeable #18

Closed SkyCol closed 1 year ago

SkyCol commented 1 year ago

Compared to AD-nerf, I found the head shaking is noticeable, is there any way have been found to improve this ?

Fictionarry commented 1 year ago

Adding --smooth_path may help when inference, which we inherit from RAD-NeRF.

SkyCol commented 1 year ago

Adding --smooth_path may help when inference, which we inherit from RAD-NeRF.

Thank you! It's helpful

SkyCol commented 1 year ago

Adding --smooth_path may help when inference, which we inherit from RAD-NeRF.

And when I do the lips finetune , I meet an error similar to https://github.com/Fictionarry/ER-NeRF/issues/13#issue-1836336970

Error: RuntimeError: Given input size: (192x3x2). Calculated output size: (192x1x0). Output size is too small

My video resolution is 476*476

Fictionarry commented 1 year ago

Adding --smooth_path may help when inference, which we inherit from RAD-NeRF.

And when I do the lips finetune , I meet an error similar to #13 (comment)

Error: RuntimeError: Given input size: (192x3x2). Calculated output size: (192x1x0). Output size is too small

My video resolution is 476*476

Well, I think it's because the lip rect is too small for LPIPS in some frames, and now I have done a fix about it, but not sure if it works for you :)

SkyCol commented 1 year ago

Adding --smooth_path may help when inference, which we inherit from RAD-NeRF.

And when I do the lips finetune , I meet an error similar to #13 (comment) Error: RuntimeError: Given input size: (192x3x2). Calculated output size: (192x1x0). Output size is too small My video resolution is 476*476

Well, I think it's because the lip rect is too small for LPIPS in some frames, and now I have done a fix about it, but not sure if it works for you :)

Well, I think it's because the lip rect is too small for LPIPS in some frames, and now I have done a fix about it, but not sure if it works for you :) Yes! the error did not appear again , thank you for your help.