search

ashawkey / RAD-NeRF

Real-time Neural Radiance Talking Portrait Synthesis via Audio-spatial Decomposition
MIT License
865 stars 151 forks source link

What is the maximum supported resolution of the model? #19

Open boolw opened 1 year ago

boolw commented 1 year ago

Is the maximum supported resolution of the model 512 512? I want to try to run 1024 768 or even higher resolution on my RTX4090, where do I need to modify the code? Or does not require any code modification?

ashawkey commented 1 year ago

@boolw No, the NeRF works with arbitrary resolution and you don't need to modify anything. However, the required GPU memory will increase, and the training/inference speed will be slower since more pixels are being renderred.

ashawkey commented 1 year ago

Also, the off-the-shelf data preprocessing code (semantic segmentation, face tracking) may run into problems, but I'm not sure.

boolw commented 1 year ago

Great, thanks for your reply

91xiaoyang commented 1 year ago

太好了，谢谢你的回复 Have you achieved it? His segmentation model doesn't seem to recognize people who are too far away

RuojiWang commented 6 months ago

@boolw No, the NeRF works with arbitrary resolution and you don't need to modify anything. However, the required GPU memory will increase, and the training/inference speed will be slower since more pixels are being renderred.

how can i get 128x128 resolution output? cause the environmental memory i deploy can only support virtual digital people with 128x128 size. thanks for advance.