NeRF-W latent appearance in coarse model

[ricshaw]

Thanks for the code! Why don't you feed the latent appearance embedding into the coarse NeRF model? Is this a mistake? In the NeRF in the Wild paper it says: "Like NeRF, we simultaneously optimize two copies of Fθ: A fine model that uses the model and losses described above, and a coarse model that uses only the latent appearance modeling component."

[kwea123]

You're correct I didn't add the appearance encoding to the coarse model, it should be a mistake. Unfortunately changing that now will break my pretrained models, so I will point the reader to that part in the readme.