{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:TT3RICPKYRS3FY4JXDC7E76GCC","short_pith_number":"pith:TT3RICPK","schema_version":"1.0","canonical_sha256":"9cf71409eac465b2e389b8c5f27fc610979ef8924945d7e16a9acd69394d601e","source":{"kind":"arxiv","id":"2112.05814","version":3},"attestation_state":"computed","paper":{"title":"Deep ViT Features as Dense Visual Descriptors","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Shai Bagon, Shir Amir, Tali Dekel, Yossi Gandelsman","submitted_at":"2021-12-10T20:15:03Z","abstract_excerpt":"We study the use of deep features extracted from a pretrained Vision Transformer (ViT) as dense visual descriptors. We observe and empirically demonstrate that such features, when extractedfrom a self-supervised ViT model (DINO-ViT), exhibit several striking properties, including: (i) the features encode powerful, well-localized semantic information, at high spatial granularity, such as object parts; (ii) the encoded semantic information is shared across related, yet different object categories, and (iii) positional bias changes gradually throughout the layers. These properties allow us to des"},"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":"2112.05814","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2021-12-10T20:15:03Z","cross_cats_sorted":[],"title_canon_sha256":"11001cdb2ba338df635c6d5d3b948215b27f441fe9c6dd1eef8148a34150f8e1","abstract_canon_sha256":"d478a83602944957b34370e53de2a8992cce62476ed58db061b463f9b876f76d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:06:53.681541Z","signature_b64":"mFbLolQzTPcT9F69WviTOfo0JKGs7YVY+893Kh2TNUDanPMuZtWLH/OhAYTgE1mKWsKM4NXmfyD011b7jluLBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9cf71409eac465b2e389b8c5f27fc610979ef8924945d7e16a9acd69394d601e","last_reissued_at":"2026-07-05T05:06:53.681071Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:06:53.681071Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Deep ViT Features as Dense Visual Descriptors","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Shai Bagon, Shir Amir, Tali Dekel, Yossi Gandelsman","submitted_at":"2021-12-10T20:15:03Z","abstract_excerpt":"We study the use of deep features extracted from a pretrained Vision Transformer (ViT) as dense visual descriptors. We observe and empirically demonstrate that such features, when extractedfrom a self-supervised ViT model (DINO-ViT), exhibit several striking properties, including: (i) the features encode powerful, well-localized semantic information, at high spatial granularity, such as object parts; (ii) the encoded semantic information is shared across related, yet different object categories, and (iii) positional bias changes gradually throughout the layers. These properties allow us to des"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2112.05814","kind":"arxiv","version":3},"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/2112.05814/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":"2112.05814","created_at":"2026-07-05T05:06:53.681128+00:00"},{"alias_kind":"arxiv_version","alias_value":"2112.05814v3","created_at":"2026-07-05T05:06:53.681128+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2112.05814","created_at":"2026-07-05T05:06:53.681128+00:00"},{"alias_kind":"pith_short_12","alias_value":"TT3RICPKYRS3","created_at":"2026-07-05T05:06:53.681128+00:00"},{"alias_kind":"pith_short_16","alias_value":"TT3RICPKYRS3FY4J","created_at":"2026-07-05T05:06:53.681128+00:00"},{"alias_kind":"pith_short_8","alias_value":"TT3RICPK","created_at":"2026-07-05T05:06:53.681128+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":16,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07230","citing_title":"`Attention-Guided Cross-Temporal Clustering for Self-Supervised Video Object Segmentation","ref_index":45,"is_internal_anchor":true},{"citing_arxiv_id":"2606.21292","citing_title":"Lightweight 3D Feature Pretraining by Bayesian Inversion of 2D Foundation Models","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29691","citing_title":"Unsupervised Semantic Segmentation Facilitates Model Understanding","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01987","citing_title":"Understanding Geometric Representations in Self-Supervised Vision Transformers via Subspace Intervention","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09608","citing_title":"TUDSR: Twice Upsampling-Diffusion for Higher Super-Resolution","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31136","citing_title":"FROST: Training-Free Few-Shot Segmentation with Frozen Features and Nonparametric Statistics","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29691","citing_title":"Unsupervised Semantic Segmentation Facilitates Model Understanding","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00514","citing_title":"Generate in Reconstruction Space, Match in Semantic Space: Transport Geometry for One-Step Generation","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20941","citing_title":"PaintCopilot: Modeling Painting as Autonomous Artistic Continuation","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16147","citing_title":"Registers Matter for Pixel-Space Diffusion Transformers","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2409.01652","citing_title":"ReKep: Spatio-Temporal Reasoning of Relational Keypoint Constraints for Robotic Manipulation","ref_index":119,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13852","citing_title":"Realiz3D: 3D Generation Made Photorealistic via Domain-Aware Learning","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10780","citing_title":"Beyond the Last Layer: Multi-Layer Representation Fusion for Visual Tokenization","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10780","citing_title":"Beyond the Last Layer: Multi-Layer Representation Fusion for Visual Tokenization","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24894","citing_title":"VISION-SLS: Safe Perception-Based Control from Learned Visual Representations via System Level Synthesis","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23670","citing_title":"Zero-Shot DINOv3-Based Image Matching via Many-to-Many Association","ref_index":1,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TT3RICPKYRS3FY4JXDC7E76GCC","json":"https://pith.science/pith/TT3RICPKYRS3FY4JXDC7E76GCC.json","graph_json":"https://pith.science/api/pith-number/TT3RICPKYRS3FY4JXDC7E76GCC/graph.json","events_json":"https://pith.science/api/pith-number/TT3RICPKYRS3FY4JXDC7E76GCC/events.json","paper":"https://pith.science/paper/TT3RICPK"},"agent_actions":{"view_html":"https://pith.science/pith/TT3RICPKYRS3FY4JXDC7E76GCC","download_json":"https://pith.science/pith/TT3RICPKYRS3FY4JXDC7E76GCC.json","view_paper":"https://pith.science/paper/TT3RICPK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2112.05814&json=true","fetch_graph":"https://pith.science/api/pith-number/TT3RICPKYRS3FY4JXDC7E76GCC/graph.json","fetch_events":"https://pith.science/api/pith-number/TT3RICPKYRS3FY4JXDC7E76GCC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TT3RICPKYRS3FY4JXDC7E76GCC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TT3RICPKYRS3FY4JXDC7E76GCC/action/storage_attestation","attest_author":"https://pith.science/pith/TT3RICPKYRS3FY4JXDC7E76GCC/action/author_attestation","sign_citation":"https://pith.science/pith/TT3RICPKYRS3FY4JXDC7E76GCC/action/citation_signature","submit_replication":"https://pith.science/pith/TT3RICPKYRS3FY4JXDC7E76GCC/action/replication_record"}},"created_at":"2026-07-05T05:06:53.681128+00:00","updated_at":"2026-07-05T05:06:53.681128+00:00"}