Jacob Munkberg

Hello @derrick-xwp The provided config https://github.com/NVlabs/nvdiffrec/blob/main/configs/nerf_lego.json for Lego was the same as used in the paper, e.g., Fig 30 here : https://nvlabs.github.io/nvdiffrec/assets/paper.pdf To improve the geometry, there are a few...

Hello @mexicantexan , 1. It depends. See above and the paper for details. For example, if the geometry is really good after the first pass, you may as well lock...

Thanks @emperor1412 , To increase texture detail, I would recommend to increase the a) texture resolution, b) rendering resolution, and c) batch size. We train a tangent space normal map...

Hello @silence401 , I need more details in order to help. In general, locking the geometry should only make things easier (I assume you learn materials and lighting jointly above)....

Nice that you got it working! Are you saying that you used an older version of nvdiffrast, and when updating to the latest version of nvdiffrast it works? Our installation...

> I still have question about the second pass result is worse than the first pass described above. It is hard to say. In the second pass, we switch from...

I'm not sure I understand the question, can you provide more details?

Yes, we didn't include the DTU experiments from our supplemental material with the MLP prediction of the SDF in the public code release. Some additional details in this issue: https://github.com/NVlabs/nvdiffrec/issues/9

No, we are not planning to release that part of the code unfortunately. Implementation details are included in Section 8.5 and 9.2 of our paper. https://nvlabs.github.io/nvdiffrec/assets/paper.pdf It is mostly the...

We do not use caffe2 in the code base, and I have not seen this error. As a first step, perhaps check if your PyTorch installation works as expected.