Closed JiatengLiu closed 4 months ago
I also do not have a particular idea. I have tried to get the normal image through gradients of sampled points in neural radiance fields. But it is a different setting.
Yes, I can only do this task in an approximate way, because ellipsoidal 3D Gaussians is not very friendly for geometry properties like normal.
Hi, I have some questions about how to transform normal map. I used a prediction network to get two normal maps of front and back, which should be opposite. I don't know how to correctly convert these normal lines under different poses to the world coordinate system, can you tell me?