Voxel carving from camera depth image

[neemoh]

I would like to carve a voxel grid from the depth map I get from an Intel RealSense camera, in python using the carve_depth_map method of open3d.geometry.VoxelGrid.

After setting the size and the center of my voxelgrid so that it would mostly encompass the corresponding point cloud of the depth image. my carved grid contains 0 voxels. I pass the intrinsic of the camera, and an identity transformation for the extrinsic. My guess is that the extrinsic is the issue. Here is the point cloud in camera coordinates.

Here is the grid if I set keep_voxels_outside_image to true to check where the grid is with respect to the point cloud:

I have tried all that came to my mind about what the extrinsic should be here, but I always get an empty grid from the carving. Any ideas would be very appreciated!