Study feasibility and approach to add depth

[mmattamala]

Depth acquisition ideas (just engineering stuff)

• Depth estimation network: Pro: easy to implement, con: more GPU overhead, depth accuracy
• LiDAR reprojection: Pro: Straightforward, accurate, con: requires more coding
• Stereo from alphasense: Pro: same hardware, con: overhead due to stereo semi-global matching, noisy, requires to use grayscale pair

Feature extraction

• Running slic on the depth image could provide interesting segmentation
• We can add an extra channel to the input image to pass it to DINO but we need to check if it's feasible

[JonasFrey96]

Merged to future ideas