How to back-project depth into camera coordinate system?

[Qianqian3764]

Hi, thanks your gread job! But I have some question：
When loading the camera's intrinsic parameters, the "pixel to ray array" is computed at the same time. Wonder why do this? I calculated the point cloud in the camera coordinate system, only using “IntM ” in the internal parameter, the result is not aligned with the other methods, I wonder if it is because "unit_ray_array_2D" is not considered.
MaGNet point clouds:

other method: