{"url":"/dataset/rel3d","name":"Rel3D","full_name":null,"description_markdown":"Understanding spatial relations (e.g., “laptop on table”) in visual input is important\r\nfor both humans and robots. Existing datasets are insufficient as they lack largescale, high-quality 3D ground truth information, which is critical for learning spatial\r\nrelations. In this paper, we fill this gap by constructing Rel3D: the first large-scale,\r\nhuman-annotated dataset for grounding spatial relations in 3D. Rel3D enables\r\nquantifying the effectiveness of 3D information in predicting spatial relations\r\non large-scale human data. Moreover, we propose minimally contrastive data\r\ncollection—a novel crowdsourcing method for reducing dataset bias. The 3D\r\nscenes in our dataset come in minimally contrastive pairs: two scenes in a pair\r\nare almost identical, but a spatial relation holds in one and fails in the other. We\r\nempirically validate that minimally contrastive examples can diagnose issues with\r\ncurrent relation detection models as well as lead to sample-efficient training. Code\r\nand data are available at https://github.com/princeton-vl/Rel3D.","description_withheld":null,"homepage":"https://github.com/princeton-vl/Rel3D","introduced_date":"2020-12-03","introduced_date_note":null,"introduced_by":{"paper":"/paper/rel3d-a-minimally-contrastive-benchmark-for-1","title":"Rel3D: A Minimally Contrastive Benchmark for Grounding Spatial Relations in 3D","first_author":"Ankit Goyal","url":null},"license":{"name":"BSD 3-Clause","url":null},"modalities":[{"name":"Images","url":"/datasets/modality/images"},{"name":"3D","url":"/datasets/modality/3d"}],"tasks":[{"name":"Spatial Relation Recognition","url":"/task/spatial-relation-recognition","datasets_with_task":"/datasets/task/spatial-relation-recognition"}],"languages":[{"name":"English","url":"/datasets/language/english"}],"variants":["Rel3D"],"data_loaders":[],"num_papers_in_archive":7,"source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28"},"benchmarks":[{"leaderboard":"/sota/spatial-relation-recognition-on-rel3d","task":"Spatial Relation Recognition","dataset_variant":"Rel3D","rows":9,"metrics":["Acc"],"first_row_in_archive_order":{"model":"Human","paper":"/paper/rel3d-a-minimally-contrastive-benchmark-for-1","metrics":{"Acc":"94.25"},"code_links":[{"title":"princeton-vl/SpatialSense","url":"https://github.com/princeton-vl/SpatialSense"},{"title":"princeton-vl/Rel3D","url":"https://github.com/princeton-vl/Rel3D"}]},"note":"rows are the archive's own order at snapshot; nothing here re-ranks them"}],"papers_with_a_benchmark_row":[{"paper":"/paper/rel3d-a-minimally-contrastive-benchmark-for-1","title":"Rel3D: A Minimally Contrastive Benchmark for Grounding Spatial Relations in 3D","date":"2020-12-03","rows_on_this_dataset":9,"code_links":2,"syntology":null}],"syntology_totals":{"read_at":"2026-09-24T18:15:14+00:00","papers_with_samples":0,"samples_harvested":0,"samples_ran":0,"samples_unverified":0,"pointer_only_for_licence":0,"papers_with_no_sample_that_ran":0,"note":"the per-paper counts above, summed; not a rate"},"papers_note":"The archive never published its papers-using-dataset list; these are papers with a leaderboard row on this dataset's benchmarks."}