{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2022:WTMT4C4F4AV4EMNMGBZJ5M7KEO","short_pith_number":"pith:WTMT4C4F","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"},"canonical_sha256":"b4d93e0b85e02bc231ac30729eb3ea23b1eea10f3756730feba603c40ea6fdb5","source":{"kind":"arxiv","id":"2201.03546","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2201.03546","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"arxiv_version","alias_value":"2201.03546v2","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2201.03546","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"pith_short_12","alias_value":"WTMT4C4F4AV4","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"WTMT4C4F4AV4EMNM","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"WTMT4C4F","created_at":"2026-05-18T12:33:33Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2022:WTMT4C4F4AV4EMNMGBZJ5M7KEO","target":"record","payload":{"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"},"canonical_sha256":"b4d93e0b85e02bc231ac30729eb3ea23b1eea10f3756730feba603c40ea6fdb5","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"},"source_kind":"arxiv","source_id":"2201.03546","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"mAKXy42eTw2xF4Br7SiKbJyt3ie6bAVKPgHa+DWgiqH7fUzy0tZgWEYjqbEgxsa92K6qXYQ20gBWDeS17qc1BQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-04T13:38:16.308581Z"},"content_sha256":"de7a96c56758be04685f311a4814d1b8716c9d50e606885939bb1f23329529fa","schema_version":"1.0","event_id":"sha256:de7a96c56758be04685f311a4814d1b8716c9d50e606885939bb1f23329529fa"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2022:WTMT4C4F4AV4EMNMGBZJ5M7KEO","target":"graph","payload":{"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"},"verdict_id":"b71a28ff-1f43-47dc-923d-048ee4939a49"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"oqMB6f5HeZEpvfKBWNoR2M+RYnT0g2Wq8YRR9k+7nedhY6iZEBWL83TKp0qJ1EuUbUbLe9w0afzRnS9ZnUKuBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-04T13:38:16.309670Z"},"content_sha256":"bc861b59ea7a11d571a18de5b878f9d666b3e08b66cb8c60ab5066a714cabf95","schema_version":"1.0","event_id":"sha256:bc861b59ea7a11d571a18de5b878f9d666b3e08b66cb8c60ab5066a714cabf95"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO/bundle.json","state_url":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-04T13:38:16Z","links":{"resolver":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO","bundle":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO/bundle.json","state":"https://pith.science/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO/state.json","well_known_bundle":"https://pith.science/.well-known/pith/WTMT4C4F4AV4EMNMGBZJ5M7KEO/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2022:WTMT4C4F4AV4EMNMGBZJ5M7KEO","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"dd8a4b1bc995781d61149e8a1705ae427fa55a3819a7bcb530e6a8e2812d504c","cross_cats_sorted":["cs.CL","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2022-01-10T18:59:10Z","title_canon_sha256":"82551c1a4c2bacd13df1784abb5b15249ffe375494248a240a31a918cd10e528"},"schema_version":"1.0","source":{"id":"2201.03546","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2201.03546","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"arxiv_version","alias_value":"2201.03546v2","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2201.03546","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"pith_short_12","alias_value":"WTMT4C4F4AV4","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"WTMT4C4F4AV4EMNM","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"WTMT4C4F","created_at":"2026-05-18T12:33:33Z"}],"graph_snapshots":[{"event_id":"sha256:bc861b59ea7a11d571a18de5b878f9d666b3e08b66cb8c60ab5066a714cabf95","target":"graph","created_at":"2026-05-17T23:38:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"LSeg achieves competitive zero-shot semantic segmentation by contrastively aligning dense pixel embeddings from a transformer with text embeddings of class labels."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"LSeg aligns per-pixel image embeddings contrastively with text label embeddings to enable zero-shot semantic segmentation."}],"snapshot_sha256":"931e15df43fde9c8d345e492bc67159bc6d2959111072035fb79ed708b16bb4e"},"formal_canon":{"evidence_count":3,"snapshot_sha256":"f174111fa4f88a18a99f42719ad6b60ecedadcb926f8d1a7455bead88c2639e6"},"paper":{"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","authors_text":"Boyi Li, Kilian Q. Weinberger, Ren\\'e Ranftl, Serge Belongie, Vladlen Koltun","cross_cats":["cs.CL","cs.LG"],"headline":"LSeg aligns per-pixel image embeddings contrastively with text label embeddings to enable zero-shot semantic segmentation.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2022-01-10T18:59:10Z","title":"Language-driven Semantic Segmentation"},"references":{"count":13,"internal_anchors":4,"resolved_work":13,"sample":[{"cited_arxiv_id":"1706.05587","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Rethinking Atrous Convolution for Semantic Image Segmentation","work_id":"6f5d4c68-8df6-4794-b125-a10bfe8d5876","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Imagenet: A large-scale hierarchical image database","work_id":"685985d1-7fbd-4ef7-9bd6-a8d79824b9be","year":2009},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Recent advances in open set recognition: A survey","work_id":"9eb68618-f32a-4c71-a302-d7a65a7fe369","year":2022},{"cited_arxiv_id":"2104.13921","doi":"","is_internal_anchor":true,"ref_index":4,"title":"Open-vocabulary Object Detection via Vision and Language Knowledge Distillation","work_id":"59541f32-18cf-4328-ad60-c9018e1401cf","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Few-shot open-set recognition using meta-learning","work_id":"7a1c13b4-e227-4fa9-8eed-0085512f8fac","year":2013}],"snapshot_sha256":"7276fb5d6e191c7fa0e4e34fbbbe06f3aacdc7db81c536158414d77ac891d0fd"},"source":{"id":"2201.03546","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-17T03:17:04.329514Z","id":"b71a28ff-1f43-47dc-923d-048ee4939a49","model_set":{"reader":"grok-4.3"},"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","pith_extraction_headline":"LSeg aligns per-pixel image embeddings contrastively with text label embeddings to enable zero-shot semantic segmentation.","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.","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."}},"verdict_id":"b71a28ff-1f43-47dc-923d-048ee4939a49"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:de7a96c56758be04685f311a4814d1b8716c9d50e606885939bb1f23329529fa","target":"record","created_at":"2026-05-17T23:38:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"dd8a4b1bc995781d61149e8a1705ae427fa55a3819a7bcb530e6a8e2812d504c","cross_cats_sorted":["cs.CL","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2022-01-10T18:59:10Z","title_canon_sha256":"82551c1a4c2bacd13df1784abb5b15249ffe375494248a240a31a918cd10e528"},"schema_version":"1.0","source":{"id":"2201.03546","kind":"arxiv","version":2}},"canonical_sha256":"b4d93e0b85e02bc231ac30729eb3ea23b1eea10f3756730feba603c40ea6fdb5","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"b4d93e0b85e02bc231ac30729eb3ea23b1eea10f3756730feba603c40ea6fdb5","first_computed_at":"2026-05-17T23:38:15.264219Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:15.264219Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"MnvsBQ1kjw9nPjsd39SYunc4QpodTgmYx8ekrezf371Y8awN3laEXwO5OGMDRVyts+d+UELiWxtClaf9E1y6DQ==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:15.264806Z","signed_message":"canonical_sha256_bytes"},"source_id":"2201.03546","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:de7a96c56758be04685f311a4814d1b8716c9d50e606885939bb1f23329529fa","sha256:bc861b59ea7a11d571a18de5b878f9d666b3e08b66cb8c60ab5066a714cabf95"],"state_sha256":"89f6cd6bf4fd289ba93637734316ae192b8bc44cc9d05ecc0cad104041a22914"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"h8yMV4YO8ROBIrHZancdu+NDhrdbOhNAb9ZRLYBJhVaoSqHgtrI95kMIhdkWESUphaJzj1pM4uA3zY3lrzaECA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-04T13:38:16.314054Z","bundle_sha256":"6324bb2630e3e1c8a805769a8ccc34417284862f1e95bef58336de1fe33f6341"}}