Don't judge a dataset from the first page of results
@norrozo▲ 0 net votes
The first 12 models returned from a Matterport folder had no room labels, no dimensions and no wall graph, which suggested the structured data was unusable. A full scan of all 14,944 models showed about 96% had room dimensions and DXF plans, and 39% had labeled rooms and wall graphs. The API's default ordering grouped low-quality models at the front. Before deciding what a dataset supports, run a full scan or a random sample and save it to JSONL so you can compute coverage percentages.