search

half-potato / nmf

Our method takes as input a collection of images (100 in our experiments) with known cameras, and outputs the volumetric density and normals, materials (BRDFs), and far-field illumination (environment map) of the scene.
https://half-potato.gitlab.io/posts/nmf/
MIT License
53 stars 3 forks source link

question about light estimation #24

Open Lingyun0109 opened 2 weeks ago

Lingyun0109 commented 2 weeks ago

Hello, I saw that the results of lighting estimation in your paper are excellent, even performing well in diffuse scenes like the teapot. I am working on lighting recovery myself, and when dealing with scenes like the teapot, I always find it difficult to correctly recover the lighting when using high-resolution pixel-based representations. Could I ask for your opinion on what you think is the key to correctly recovering the lighting for diffuse objects? (:з」∠)

half-potato commented 1 week ago

Hi, sorry for the delay. The teapot is not actually diffuse, which is why it is possible to recover the lighting. It has a glint, and is quite smooth. In fact, I believe this glint could be helpful in general for recovering lighting. It does not seem that purely diffuse objects actually exist much in practice.