{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:VNWXVILVH6SRGJYXD3IOX23Y3A","short_pith_number":"pith:VNWXVILV","schema_version":"1.0","canonical_sha256":"ab6d7aa1753fa51327171ed0ebeb78d8255426ce16dd3da77de55bb13fbf6019","source":{"kind":"arxiv","id":"2605.22658","version":1},"attestation_state":"computed","paper":{"title":"SegCompass: Exploring Interpretable Alignment with Sparse Autoencoders for Enhanced Reasoning Segmentation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","cs.MM","eess.IV"],"primary_cat":"cs.CV","authors_text":"Haoqian Kang, Jinpeng Wang, Ke Chen, Liupeng Li, Yan Feng, Yaowei Wang, Zhenyu Lu","submitted_at":"2026-05-21T15:59:39Z","abstract_excerpt":"While large language models provide strong compositional reasoning, existing reasoning segmentation pipelines fail to transparently connect this reasoning to visual perception. Current methods, such as latent query alignment, are end-to-end yet opaque \"black boxes\". Conversely, textual localization readout is merely readable, not truly interpretable, often functioning as an unconstrained post-hoc step. To bridge this interpretability gap, we propose SegCompass, an end-to-end model that leverages a Sparse Autoencoder (SAE) to forge an explicit, interpretable, and differentiable alignment pathwa"},"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":"2605.22658","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-21T15:59:39Z","cross_cats_sorted":["cs.LG","cs.MM","eess.IV"],"title_canon_sha256":"1cfc425f2305bd242d3ffdef8dbb7baa1f9fe8d7bd1ec608cfe918239c37256d","abstract_canon_sha256":"8c3f36555b2de0a4eee202befd475910b8ee8031e6fa74a21537865d52717113"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-22T02:04:48.647494Z","signature_b64":"bagR6XAtZH+T5twVv0W3Z1x+PXn9Zr7rYOuiu2PQbaD9FZ3dGGxwFuztfO4BW/mNFutcNpXG7xCUh3PSGU4ABQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ab6d7aa1753fa51327171ed0ebeb78d8255426ce16dd3da77de55bb13fbf6019","last_reissued_at":"2026-05-22T02:04:48.646711Z","signature_status":"signed_v1","first_computed_at":"2026-05-22T02:04:48.646711Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SegCompass: Exploring Interpretable Alignment with Sparse Autoencoders for Enhanced Reasoning Segmentation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","cs.MM","eess.IV"],"primary_cat":"cs.CV","authors_text":"Haoqian Kang, Jinpeng Wang, Ke Chen, Liupeng Li, Yan Feng, Yaowei Wang, Zhenyu Lu","submitted_at":"2026-05-21T15:59:39Z","abstract_excerpt":"While large language models provide strong compositional reasoning, existing reasoning segmentation pipelines fail to transparently connect this reasoning to visual perception. Current methods, such as latent query alignment, are end-to-end yet opaque \"black boxes\". Conversely, textual localization readout is merely readable, not truly interpretable, often functioning as an unconstrained post-hoc step. To bridge this interpretability gap, we propose SegCompass, an end-to-end model that leverages a Sparse Autoencoder (SAE) to forge an explicit, interpretable, and differentiable alignment pathwa"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.22658","kind":"arxiv","version":1},"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/2605.22658/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":"2605.22658","created_at":"2026-05-22T02:04:48.646853+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.22658v1","created_at":"2026-05-22T02:04:48.646853+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.22658","created_at":"2026-05-22T02:04:48.646853+00:00"},{"alias_kind":"pith_short_12","alias_value":"VNWXVILVH6SR","created_at":"2026-05-22T02:04:48.646853+00:00"},{"alias_kind":"pith_short_16","alias_value":"VNWXVILVH6SRGJYX","created_at":"2026-05-22T02:04:48.646853+00:00"},{"alias_kind":"pith_short_8","alias_value":"VNWXVILV","created_at":"2026-05-22T02:04:48.646853+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.23655","citing_title":"CVSearch: Empowering Multimodal LLMs with Cognitive Visual Search for High-Resolution Image Perception","ref_index":7,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VNWXVILVH6SRGJYXD3IOX23Y3A","json":"https://pith.science/pith/VNWXVILVH6SRGJYXD3IOX23Y3A.json","graph_json":"https://pith.science/api/pith-number/VNWXVILVH6SRGJYXD3IOX23Y3A/graph.json","events_json":"https://pith.science/api/pith-number/VNWXVILVH6SRGJYXD3IOX23Y3A/events.json","paper":"https://pith.science/paper/VNWXVILV"},"agent_actions":{"view_html":"https://pith.science/pith/VNWXVILVH6SRGJYXD3IOX23Y3A","download_json":"https://pith.science/pith/VNWXVILVH6SRGJYXD3IOX23Y3A.json","view_paper":"https://pith.science/paper/VNWXVILV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.22658&json=true","fetch_graph":"https://pith.science/api/pith-number/VNWXVILVH6SRGJYXD3IOX23Y3A/graph.json","fetch_events":"https://pith.science/api/pith-number/VNWXVILVH6SRGJYXD3IOX23Y3A/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VNWXVILVH6SRGJYXD3IOX23Y3A/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VNWXVILVH6SRGJYXD3IOX23Y3A/action/storage_attestation","attest_author":"https://pith.science/pith/VNWXVILVH6SRGJYXD3IOX23Y3A/action/author_attestation","sign_citation":"https://pith.science/pith/VNWXVILVH6SRGJYXD3IOX23Y3A/action/citation_signature","submit_replication":"https://pith.science/pith/VNWXVILVH6SRGJYXD3IOX23Y3A/action/replication_record"}},"created_at":"2026-05-22T02:04:48.646853+00:00","updated_at":"2026-05-22T02:04:48.646853+00:00"}