search

PeterouZh / CIPS-3D

3D-aware GANs based on NeRF (arXiv).
MIT License
610 stars 60 forks source link

How to test on my own image datasets? #8

Closed qian5683 closed 2 years ago

qian5683 commented 2 years ago

How to test on my own image datasets?

PeterouZh commented 2 years ago

I'm not sure what you mean. Do you want to train on your own dataset?

qian5683 commented 2 years ago

I'm not sure what you mean. Do you want to train on your own dataset?

I want to test the model with other pictures, but I don't find a way to choose pictures

PeterouZh commented 2 years ago

If you want to test on a real image, you first need to do GAN inversion to project the image into the w space.

However, the inversion method for 3D GANs is not yet mature and robust. For example, it has a poor effect on the profile face with an unknown pose.

I have no idea how to deal with this problem now. If anyone can provide a reasonable idea, maybe I can implement this function quickly. However, currently, the program does not support this feature.