Camera extrinsics, intrinsics and depth

[anilesec]

The intrinsics/xxxxxxxxx_640x480.yml file contains intrinsics separately for 'color' and 'depth'. However, to convert the depth map to pointcloud in SequenceLoader() 'color' intrinsics are used here

Isn't intr['depth'] be used for converting depth map into pointlcouds? what is the purpose of 'depth' intrinsics?

Also, the pointclouds return here are they world coordinate system?

[ychao-nvidia]

The recorded depth images have been aligned to the color image space. So we should be using the color intrinsics for depth.

Depth intrinsics does not have any use here.

Yes, point clouds are in world coordinates. And the world coordinates here is actually one of the camera frame. So this is just transforming all the point clouds to one particular camera's frame.