Now with a “hide-machine” feature that allows to see what’s below the Lowrider beam by remembering the last non-obstructed pixels and masking accordingly.
https://youtu.be/MeI6lupNXM4?si=lk8JDMuJgv3LqsoJ
Currently using the Depth Anything V2 depth to estimate depth. Then assumes the largest depth cluster is the table and masks everything above a configurable offset. Works reasonably well.