{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:WTMT4C4F4AV4EMNMGBZJ5M7KEO","short_pith_number":"pith:WTMT4C4F","schema_version":"1.0","canonical_sha256":"b4d93e0b85e02bc231ac30729eb3ea23b1eea10f3756730feba603c40ea6fdb5","source":{"kind":"arxiv","id":"2201.03546","version":2},"attestation_state":"computed","paper":{"title":"Language-driven Semantic Segmentation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"LSeg aligns per-pixel image embeddings contrastively with text label embeddings to enable zero-shot semantic segmentation.","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CV","authors_text":"Boyi Li, Kilian Q. Weinberger, Ren\\'e Ranftl, Serge Belongie, Vladlen Koltun","submitted_at":"2022-01-10T18:59:10Z","abstract_excerpt":"We present LSeg, a novel model for language-driven semantic image segmentation. LSeg uses a text encoder to compute embeddings of descriptive input labels (e.g., \"grass\" or \"building\") together with a transformer-based image encoder that computes dense per-pixel embeddings of the input image. The image encoder is trained with a contrastive objective to align pixel embeddings to the text embedding of the corresponding semantic class. The text embeddings provide a flexible label representation in which semantically similar labels map to similar regions in the embedding space (e.g., \"cat\" and \"fu"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2201.03546","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2022-01-10T18:59:10Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"82551c1a4c2bacd13df1784abb5b15249ffe375494248a240a31a918cd10e528","abstract_canon_sha256":"dd8a4b1bc995781d61149e8a1705ae427fa55a3819a7bcb530e6a8e2812d504c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:15.264806Z","signature_b64":"MnvsBQ1kjw9nPjsd39SYunc4QpodTgmYx8ekrezf371Y8awN3laEXwO5OGMDRVyts+d+UELiWxtClaf9E1y6DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b4d93e0b85e02bc231ac30729eb3ea23b1eea10f3756730feba603c40ea6fdb5","last_reissued_at":"2026-05-17T23:38:15.264219Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:15.264219Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Language-driven Semantic Segmentation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"LSeg aligns per-pixel image embeddings contrastively with text label embeddings to enable zero-shot semantic segmentation.","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CV","authors_text":"Boyi Li, Kilian Q. Weinberger, Ren\\'e Ranftl, Serge Belongie, Vladlen Koltun","submitted_at":"2022-01-10T18:59:10Z","abstract_excerpt":"We present LSeg, a novel model for language-driven semantic image segmentation. LSeg uses a text encoder to compute embeddings of descriptive input labels (e.g., \"grass\" or \"building\") together with a transformer-based image encoder that computes dense per-pixel embeddings of the input image. The image encoder is trained with a contrastive objective to align pixel embeddings to the text embedding of the corresponding semantic class. The text embeddings provide a flexible label representation in which semantically similar labels map to similar regions in the embedding space (e.g., \"cat\" and \"fu"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We demonstrate that our approach achieves highly competitive zero-shot performance compared to existing zero- and few-shot semantic segmentation methods, and even matches the accuracy of traditional segmentation algorithms when a fixed label set is provided.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the contrastive alignment learned on seen classes will transfer to arbitrary unseen text labels without retraining or additional samples, relying on the semantic structure already present in the pre-trained text encoder.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LSeg achieves competitive zero-shot semantic segmentation by contrastively aligning dense pixel embeddings from a transformer with text embeddings of class labels.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"LSeg aligns per-pixel image embeddings contrastively with text label embeddings to enable zero-shot semantic segmentation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"931e15df43fde9c8d345e492bc67159bc6d2959111072035fb79ed708b16bb4e"},"source":{"id":"2201.03546","kind":"arxiv","version":2},"verdict":{"id":"b71a28ff-1f43-47dc-923d-048ee4939a49","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T03:17:04.329514Z","strongest_claim":"We demonstrate that our approach achieves highly competitive zero-shot performance compared to existing zero- and few-shot semantic segmentation methods, and even matches the accuracy of traditional segmentation algorithms when a fixed label set is provided.","one_line_summary":"LSeg achieves competitive zero-shot semantic segmentation by contrastively aligning dense pixel embeddings from a transformer with text embeddings of class labels.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the contrastive alignment learned on seen classes will transfer to arbitrary unseen text labels without retraining or additional samples, relying on the semantic structure already present in the pre-trained text encoder.","pith_extraction_headline":"LSeg aligns per-pixel image embeddings contrastively with text label embeddings to enable zero-shot semantic segmentation."},"references":{"count":13,"sample":[{"doi":"","year":null,"title":"Rethinking Atrous Convolution for Semantic Image Segmentation","work_id":"6f5d4c68-8df6-4794-b125-a10bfe8d5876","ref_index":1,"cited_arxiv_id":"1706.05587","is_internal_anchor":true},{"doi":"","year":2009,"title":"Imagenet: A large-scale hierarchical image database","work_id":"685985d1-7fbd-4ef7-9bd6-a8d79824b9be","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Recent advances in open set recognition: A survey","work_id":"9eb68618-f32a-4c71-a302-d7a65a7fe369","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Open-vocabulary Object Detection via Vision and Language Knowledge Distillation","work_id":"59541f32-18cf-4328-ad60-c9018e1401cf","ref_index":4,"cited_arxiv_id":"2104.13921","is_internal_anchor":true},{"doi":"","year":2013,"title":"Few-shot open-set recognition using meta-learning","work_id":"7a1c13b4-e227-4fa9-8eed-0085512f8fac","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":13,"snapshot_sha256":"7276fb5d6e191c7fa0e4e34fbbbe06f3aacdc7db81c536158414d77ac891d0fd","internal_anchors":4},"formal_canon":{"evidence_count":3,"snapshot_sha256":"f174111fa4f88a18a99f42719ad6b60ecedadcb926f8d1a7455bead88c2639e6"},"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":"2201.03546","created_at":"2026-05-17T23:38:15.264326+00:00"},{"alias_kind":"arxiv_version","alias_value":"2201.03546v2","created_at":"2026-05-17T23:38:15.264326+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2201.03546","created_at":"2026-05-17T23:38:15.264326+00:00"},{"alias_kind":"pith_short_12","alias_value":"WTMT4C4F4AV4","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_16","alias_value":"WTMT4C4F4AV4EMNM","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_8","alias_value":"WTMT4C4F","created_at":"2026-05-18T12:33:33.725879+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":30,"internal_anchor_count":30,"sample":[{"citing_arxiv_id":"2605.06088","citing_title":"OpenGaFF: Open-Vocabulary Gaussian Feature Field with Codebook Attention","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2605.23287","citing_title":"LangFlash: Feed-forward 3D Language Gaussian Splatting from Sparse Unposed Images","ref_index":19,"is_internal_anchor":true},{"citing_arxiv_id":"2605.23188","citing_title":"SpikingMoE: SDPrompt-Guided Dynamic Expert Fusion in Spiking Neural Networks","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2410.10247","citing_title":"LPT: Less-overfitting Prompt Tuning for Vision-Language Model","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17630","citing_title":"SegRAG: Training-Free Retrieval-Augmented Semantic Segmentation","ref_index":33,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06088","citing_title":"OpenGaFF: Open-Vocabulary Gaussian Feature Field with Codebook Attention","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2605.13161","citing_title":"A$_3$B$_2$: Adaptive Asymmetric Adapter for Alleviating Branch Bias in Vision-Language Image Classification with Few-Shot Learning","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2605.15942","citing_title":"Decomposed Vision-Language Alignment for Fine-Grained Open-Vocabulary Segmentation","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17630","citing_title":"SegRAG: Training-Free Retrieval-Augmented Semantic Segmentation","ref_index":33,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20044","citing_title":"OP2GS: Object-Aware 3D Gaussian Splatting with Dual-Opacity Primitives","ref_index":24,"is_internal_anchor":true},{"citing_arxiv_id":"2605.16903","citing_title":"WOW-Seg: A Word-free Open World Segmentation Model","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2511.13312","citing_title":"EL3DD: Extended Latent 3D Diffusion for Language Conditioned Multitask Manipulation","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2511.19704","citing_title":"RADSeg: Unleashing Parameter and Compute Efficient Zero-Shot Open-Vocabulary Segmentation Using Agglomerative Models","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2512.04021","citing_title":"C3G: Learning Compact 3D Representations with 2K Gaussians","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2512.08730","citing_title":"SegEarth-OV3: Exploring SAM 3 for Open-Vocabulary Semantic Segmentation in Remote Sensing Images","ref_index":35,"is_internal_anchor":true},{"citing_arxiv_id":"2512.17541","citing_title":"FLEG: Feed-Forward Language Embedded Gaussian Splatting from Any Views via Compact Semantic Representation","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2603.20530","citing_title":"Memory Over Maps: 3D Object Localization Without Reconstruction","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2605.13161","citing_title":"A$_3$B$_2$: Adaptive Asymmetric Adapter for Alleviating Branch Bias in Vision-Language Image Classification with Few-Shot Learning","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2604.03309","citing_title":"TreeGaussian: Tree-Guided Cascaded Contrastive Learning for Hierarchical Consistent 3D Gaussian Scene Segmentation and Understanding","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2604.01204","citing_title":"Neural Harmonic Textures for High-Quality Primitive Based Neural Reconstruction","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11771","citing_title":"Revisiting Shadow Detection from a Vision-Language Perspective","ref_index":57,"is_internal_anchor":true},{"citing_arxiv_id":"2604.26221","citing_title":"Seeking Consensus: Geometric-Semantic On-the-Fly Recalibration for Open-Vocabulary Remote Sensing Semantic Segmentation","ref_index":19,"is_internal_anchor":true},{"citing_arxiv_id":"2604.24575","citing_title":"Diffusion Model as a Generalist Segmentation Learner","ref_index":51,"is_internal_anchor":true},{"citing_arxiv_id":"2604.22439","citing_title":"NRGS: Neural Regularization for Robust 3D Semantic Gaussian Splatting","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06088","citing_title":"OpenGaFF: Open-Vocabulary Gaussian Feature Field with Codebook Attention","ref_index":12,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":3,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO","json":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO.json","graph_json":"https://pith.science/api/pith-number/WTMT4C4F4AV4EMNMGBZJ5M7KEO/graph.json","events_json":"https://pith.science/api/pith-number/WTMT4C4F4AV4EMNMGBZJ5M7KEO/events.json","paper":"https://pith.science/paper/WTMT4C4F"},"agent_actions":{"view_html":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO","download_json":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO.json","view_paper":"https://pith.science/paper/WTMT4C4F","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2201.03546&json=true","fetch_graph":"https://pith.science/api/pith-number/WTMT4C4F4AV4EMNMGBZJ5M7KEO/graph.json","fetch_events":"https://pith.science/api/pith-number/WTMT4C4F4AV4EMNMGBZJ5M7KEO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO/action/storage_attestation","attest_author":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO/action/author_attestation","sign_citation":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO/action/citation_signature","submit_replication":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO/action/replication_record"}},"created_at":"2026-05-17T23:38:15.264326+00:00","updated_at":"2026-05-17T23:38:15.264326+00:00"}