{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:HI7OBLFCMDLGNJHEYDCW4YEMCK","short_pith_number":"pith:HI7OBLFC","schema_version":"1.0","canonical_sha256":"3a3ee0aca260d666a4e4c0c56e608c1281e84374354ab3cd3db0299f6dec23d4","source":{"kind":"arxiv","id":"2503.13111","version":2},"attestation_state":"computed","paper":{"title":"MM-Spatial: Exploring 3D Spatial Understanding in Multimodal LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CV","authors_text":"Afshin Dehghan, David Griffiths, Erik Daxberger, Gefen Kohavi, Haiming Gang, Justin Lazarow, Kai Kang, Marcin Eichner, Nina Wenzel, Peter Grasch, Yinfei Yang","submitted_at":"2025-03-17T12:34:22Z","abstract_excerpt":"Multimodal large language models (MLLMs) excel at 2D visual understanding but remain limited in their ability to reason about 3D space. In this work, we leverage large-scale high-quality 3D scene data with open-set annotations to introduce 1) a novel supervised fine-tuning dataset and 2) a new evaluation benchmark, focused on indoor scenes. Our Cubify Anything VQA (CA-VQA) data covers diverse spatial tasks including spatial relationship prediction, metric size and distance estimation, and 3D grounding. We show that CA-VQA enables us to train MM-Spatial, a strong generalist MLLM that also achie"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2503.13111","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-03-17T12:34:22Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"2383193acf61fcea15f4cff1fdc5cb27881882dc1bd85f3f053c6279effbbe39","abstract_canon_sha256":"0642e52074cd9a9edfe00b244ad4161c371bc728efe9a2d2b558a624be4d1f34"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:06:08.180013Z","signature_b64":"n06uhTbiPrtympoh59piv5FPcUma62VwtxUAEN8/UtfbMnULpX2h8UeuwjD5Y+94QDoWoxUFWaSMRzLQdTwzDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3a3ee0aca260d666a4e4c0c56e608c1281e84374354ab3cd3db0299f6dec23d4","last_reissued_at":"2026-07-05T12:06:08.179507Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:06:08.179507Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MM-Spatial: Exploring 3D Spatial Understanding in Multimodal LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CV","authors_text":"Afshin Dehghan, David Griffiths, Erik Daxberger, Gefen Kohavi, Haiming Gang, Justin Lazarow, Kai Kang, Marcin Eichner, Nina Wenzel, Peter Grasch, Yinfei Yang","submitted_at":"2025-03-17T12:34:22Z","abstract_excerpt":"Multimodal large language models (MLLMs) excel at 2D visual understanding but remain limited in their ability to reason about 3D space. In this work, we leverage large-scale high-quality 3D scene data with open-set annotations to introduce 1) a novel supervised fine-tuning dataset and 2) a new evaluation benchmark, focused on indoor scenes. Our Cubify Anything VQA (CA-VQA) data covers diverse spatial tasks including spatial relationship prediction, metric size and distance estimation, and 3D grounding. We show that CA-VQA enables us to train MM-Spatial, a strong generalist MLLM that also achie"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.13111","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2503.13111/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2503.13111","created_at":"2026-07-05T12:06:08.179565+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.13111v2","created_at":"2026-07-05T12:06:08.179565+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.13111","created_at":"2026-07-05T12:06:08.179565+00:00"},{"alias_kind":"pith_short_12","alias_value":"HI7OBLFCMDLG","created_at":"2026-07-05T12:06:08.179565+00:00"},{"alias_kind":"pith_short_16","alias_value":"HI7OBLFCMDLGNJHE","created_at":"2026-07-05T12:06:08.179565+00:00"},{"alias_kind":"pith_short_8","alias_value":"HI7OBLFC","created_at":"2026-07-05T12:06:08.179565+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23717","citing_title":"SpaCE: Rethinking Spatial Capacity and Generalization in Multi-Frame Multimodal Large Language Models","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31257","citing_title":"Decodable Is Not Grounded: A Vision-Ablation Arbiter for VLM Spatial Reasoning","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2601.22228","citing_title":"Lost in Space? Vision-Language Models Struggle with Relative Camera Pose Estimation","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.00799","citing_title":"Multimodal Language Models Cannot Spot Spatial Inconsistencies","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26934","citing_title":"World2VLM: Distilling World Model Imagination into VLMs for Dynamic Spatial Reasoning","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HI7OBLFCMDLGNJHEYDCW4YEMCK","json":"https://pith.science/pith/HI7OBLFCMDLGNJHEYDCW4YEMCK.json","graph_json":"https://pith.science/api/pith-number/HI7OBLFCMDLGNJHEYDCW4YEMCK/graph.json","events_json":"https://pith.science/api/pith-number/HI7OBLFCMDLGNJHEYDCW4YEMCK/events.json","paper":"https://pith.science/paper/HI7OBLFC"},"agent_actions":{"view_html":"https://pith.science/pith/HI7OBLFCMDLGNJHEYDCW4YEMCK","download_json":"https://pith.science/pith/HI7OBLFCMDLGNJHEYDCW4YEMCK.json","view_paper":"https://pith.science/paper/HI7OBLFC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.13111&json=true","fetch_graph":"https://pith.science/api/pith-number/HI7OBLFCMDLGNJHEYDCW4YEMCK/graph.json","fetch_events":"https://pith.science/api/pith-number/HI7OBLFCMDLGNJHEYDCW4YEMCK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HI7OBLFCMDLGNJHEYDCW4YEMCK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HI7OBLFCMDLGNJHEYDCW4YEMCK/action/storage_attestation","attest_author":"https://pith.science/pith/HI7OBLFCMDLGNJHEYDCW4YEMCK/action/author_attestation","sign_citation":"https://pith.science/pith/HI7OBLFCMDLGNJHEYDCW4YEMCK/action/citation_signature","submit_replication":"https://pith.science/pith/HI7OBLFCMDLGNJHEYDCW4YEMCK/action/replication_record"}},"created_at":"2026-07-05T12:06:08.179565+00:00","updated_at":"2026-07-05T12:06:08.179565+00:00"}