Calrification on derivative variable names

[VladimirYugay]

Hey there,

1. In backward.cu dL_dtx, dL_dty, dL_dtz are derivatives w.r.t. Gaussian means in the camera space?
2. In backward.cu when I run glm::mat3 T = W * J; with the debugger, it gives me the same result if I'd do T = J @ W in numpy. I read that in opengl matrices I column-major, but still it's a bit unclear.
3. In backward.cu dL_dTij are derivatives w.r.t. T = W * J or T = J * W in "mathematical"/"theoretical" sense?

[yyzzyy78]

Hi, have you figured it out? I found the same result about: glm::mat3 T = W * J; In fact, I have the result of T=J@W, corresponding to the matmul.

[VladimirYugay]

@yyzzyy78 check out this repo, they actually implemented it if the gaussians are frozen!

[qpc001]

Hey there,

1. In backward.cu dL_dtx, dL_dty, dL_dtz are derivatives w.r.t. Gaussian means in the camera space?
2. In backward.cu when I run glm::mat3 T = W * J; with the debugger, it gives me the same result if I'd do T = J @ W in numpy. I read that in opengl matrices I column-major, but still it's a bit unclear.
3. In backward.cu dL_dTij are derivatives w.r.t. T = W * J or T = J * W in "mathematical"/"theoretical" sense?

嘿，这里，

1. 在backward.cu``dL_dtx, dL_dty, dL_dtz中是w.r.t.的导数。高斯意味着在相机空间中？ 在
2. backward.cu中，当我用调试器运行glm::mat3 T = W * J;时，如果我在numpy中运行T = J @ W，它会给我同样的结果。我在opengl matrices中读到过，但还是有点不清楚。 在
3. backward.cu``dL_dTij中，是关于t的导数。T = W * J或T = J * W在“数学”/“理论”意义上？

The glm::mat is a transpose for mat input, which is store in col.