Closed tianxiaguixin002 closed 1 month ago
Hi, L539 is -r_g and L542 is the element-wise operation. And then we normalize it in L547 to form a normal in the local gaussian coordinate system. After that, we transform/rotate it to the view space (first to world space then to current view coordinate with inverse of view2gaussian) in L550. In the paper, we transform the normal the world space and ignore the normalisation for simplicity. This two processes is equivalent.
Thank you !
Thanks to the author for open source this great work. I'm confused about calculating the normal of a point in the code:
in the parer, point normal is computed in following formula:![point-compute2](https://github.com/autonomousvision/gaussian-opacity-fields/assets/26108782/37da5f9e-60c5-4763-b7aa-d0a627f708db)
There is a big gap between the formula in the paper and the implementation in the code. Can you explain it in more detail? Thank you very much.