search

elliottwu / unsup3d

(CVPR'20 Oral) Unsupervised Learning of Probably Symmetric Deformable 3D Objects from Images in the Wild
MIT License
1.19k stars 192 forks source link

Higher input resolution? #5

Closed puppet101 closed 4 years ago

puppet101 commented 4 years ago

Hi, the input face size is 64*64 which is too small, have you tried higher resolution for better image quality?

elliottwu commented 4 years ago

Hi! We did try 128x128 at some point, but I do not have a completed model at the moment. It should be quite straightforward to add one more downsampling/upsampling layer to the encoder/decoder networks. But training will be ~4 times slower because of the rendering. And you might also need to adjust the perceptual loss, eg, using a deeper layer.