The Dataset Viewer has been disabled on this dataset.
Revisit artifacts
Intermediate outputs of Revisit on the
SceneDiff benchmark (MIT), so the SceneDiff table and the
verifier training can be reproduced without recomputing everything. python scripts/download_data.py --artifacts
downloads and unpacks them into data/artifacts/.
| Archive | Unpacks to | Contents |
|---|---|---|
scenediff_proposals.tar |
scenediff/proposals/<scene>/ |
Stage-1 proposals (Pi3 + DINOv3 ViT-H+/16 + SAM 1) for the 105 validation and 249 test scenes: object_masks.pkl, verifier_cards.json, evidence crops, oracle_gt.json |
scenediff_verifier_outputs.tar |
scenediff/verifier_outputs/<verifier>/<scene>/ |
per-proposal answers (verifier_results.json) and filtered object_masks.pkl for every row of the SceneDiff table |
card_pool.tar |
card_pool/<variant>/<scene>/ |
four more Stage-1 variants on the validation scenes, used only to train the room adapters |
One card-pool variant (pi3x_dinov3-vith16plus_sam3) was produced with a bug that placed SAM 3 point prompts
outside the image; it is kept because the released room adapters were trained on it.
living_room_19_living_room_20 (test) has no Stage-1 output, so the official evaluator scores 249 test scenes.
- Downloads last month
- 25