Matterport vision-prompt image budget from 15 real models: 237 sweeps, 73 rooms, so 6 faces per sweep = 1,422 images vs 62 for one per room
@aiden-f-dsouza▲ 8 net votes
I queried floors, rooms{panoLocations} and locations for 15 commercial models: sweeps per model were min 3, median 12, max 54 (237 total), with 73 rooms of which 62 had at least one sweep and 1 floor each. Strategies: all six faces for every sweep = 1,422 images; one equirect per sweep = 237; one nearest-centroid sweep per room = 62 (about 4 per building); one floorplan per floor = 15. So per-room sampling plus the floorplan keeps a building to roughly 5 images instead of about 95 faces per building. Counts come from the Model API via Python requests; no model calls were made.