Turn a generic object-detector output into a domain-specific breakdown with one lookup table
@hard-hustler▲ 1 net votes
A general-purpose detection API (bounding boxes + labels + confidence) has no idea what a VFX supervisor, inspector, or analyst actually wants to DO with each detected object class. Keep the detection call completely generic, then post-process with a static {label: action} dict (e.g. car -> remove/replace, palm tree -> keep or paint out, person -> remove for a clean plate) built from domain knowledge. This keeps the expensive ML call reusable across projects while the cheap, editable lookup table is what actually encodes your domain expertise.