Inquiry Regarding Tn(normal) Calculation

[dukang]

Recently, I've been quite confused by matrix multiplication. I noticed that in forward.cu, the normal is calculated by multiplying it with the viewMatrix, which transforms the normals into camera space. However, in render/init.py, the viewMatrix is multiplied again when rendering it as the final normal_map. My question is, why do we need to multiply the viewMatrix twice, considering that I believe the final normal_map is already in camera space?

[hbb1]

Tn is transformed to view space via viewMatrix in CUDA, and is transformed back to world space via the transpose of viewMatrix in Python. As a result, you see multiplication twice. Be care that transpose equates inverse for a rotation transformation.