[Feature request] Add `depth_to_disparity` and `disparity_to_depth` methods in `StereoCamera` class

🚀 Feature

Two methods to convert between disparity and depth maps. This should be batched and be general, i.e. the normal conversion depth = (baseline * focal length) / disparity) doesn’t work for different focal lengths.

Motivation

Some downstream tasks in 3D vision uses depth maps rather than point clouds or disparity maps.

Pitch

Add two new methods:

• disparity_to_depth
• depth_to_disparity

To the StereoCamera class. Both methods should take as input a tensor of shape (B, H, W, 1) and return the same shape. Core logic should be separated into its own function and there should be extensive tests. Tests can utilize already added test data in the test_stereo.py module.

[nitaifingerhut]

I can also guide you through this or maybe do it myself if I’ll find the time

[copaah]

@ankitajais20

Do you still want to contribute to this? I can supervise you if you have any questions - just ping me here or on slack.