kuldeepbrd1 / diff-3dgs-cam-grad Goto Github PK

Hi,

Thank you for your work. I think your work is really helpful to my project. I saw your work in the repo of 3dgs, and I want to ask did you proved that your code work can work? I noticed that you let others to check if your code is reasonable five days ago. So I am not sure whether I can directly use your code in my project. Btw, I also want to ask if use your code, what should i change in 3d-gs code. Thank you.

Best,
Kevin

Can you please elaborate on the gradient computation starting here?

While this makes sense:

// dL/dW = dL/dT *dT/dW
// T = W * J
// dL/dW = J^T * dL/dT

You mentioned this paper, and there it seems that dW/dTcw is a 9x6 matrix, and I'm not sure how should it match dl/dW of shape (3x3) that you get.

Also, shouldn't dl_dw look like this?

float dL_dW00 = J[0][0] * dL_dT00 + J[1][0] * dL_dT10;
float dl_dW01 = J[0][0] * dL_dT01 + J[1][0] * dL_dT11;
float dl_dW02 = J[0][0] * dL_dT02 + J[1][0] * dL_dT12;
float dl_dW10 = J[0][1] * dL_dT00 + J[1][1] * dL_dT10;
float dl_dW11 = J[0][1] * dL_dT01 + J[1][1] * dL_dT11;
float dl_dW12 = J[0][1] * dL_dT02 + J[1][1] * dL_dT12;
float dl_dW20 = J[0][2] * dL_dT00 + J[1][2] * dL_dT10;
float dl_dW21 = J[0][2] * dL_dT01 + J[1][2] * dL_dT11;
float dl_dW22 = J[0][2] * dL_dT02 + J[1][2] * dL_dT12;