{"paper":{"title":"GraSP-VL: Length as a Semantic Granularity Interface for Vision-Language Representations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A learned prefix transform reorganizes frozen vision-language embeddings so that length directly controls semantic granularity from coarse to fine.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Chengchang Pan, Honggang Qi, Zesheng Li","submitted_at":"2026-05-18T01:10:07Z","abstract_excerpt":"Frozen vision-language embeddings contain signals at multiple semantic resolutions, from object identity to attributes, relations, and full-caption meaning, but they expose these signals through a fixed-length vector interface. We study whether embedding length can be turned into a controllable semantic access interface. We propose \\textbf{GraSP-VL}, which learns a shared near-orthogonal prefix transform over frozen VLM embeddings. GraSP-VL instantiates a \\textbf{Semantic Matryoshka} interface: short prefixes are assigned coarse semantic roles, while longer prefixes progressively expose finer "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"frozen VLM embeddings can be reorganized into a truncatable semantic prefix interface rather than merely compressed","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"A single shared near-orthogonal prefix transform learned over frozen embeddings can progressively assign coarse-to-fine semantic roles to increasing prefix lengths while preserving full-dimensional geometry (stated in the abstract description of the Semantic Matryoshka interface).","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GraSP-VL turns frozen VLM embedding length into a controllable semantic granularity interface via a learned shared prefix transform that creates a Semantic Matryoshka structure.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A learned prefix transform reorganizes frozen vision-language embeddings so that length directly controls semantic granularity from coarse to fine.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0ae9466257fe2c225774cc7549cea7c1ae8bbb42b071d2e7e3284238293be069"},"source":{"id":"2605.17727","kind":"arxiv","version":1},"verdict":{"id":"0b3c2a24-b2dd-4b02-83bc-857e7394a218","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T21:46:26.045362Z","strongest_claim":"frozen VLM embeddings can be reorganized into a truncatable semantic prefix interface rather than merely compressed","one_line_summary":"GraSP-VL turns frozen VLM embedding length into a controllable semantic granularity interface via a learned shared prefix transform that creates a Semantic Matryoshka structure.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"A single shared near-orthogonal prefix transform learned over frozen embeddings can progressively assign coarse-to-fine semantic roles to increasing prefix lengths while preserving full-dimensional geometry (stated in the abstract description of the Semantic Matryoshka interface).","pith_extraction_headline":"A learned prefix transform reorganizes frozen vision-language embeddings so that length directly controls semantic granularity from coarse to fine."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.17727/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T22:01:19.418180Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T21:50:54.848925Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"shingle_duplication","ran_at":"2026-05-19T21:49:43.460598Z","status":"completed","version":"0.1.0","findings_count":0},{"name":"citation_quote_validity","ran_at":"2026-05-19T21:49:43.289724Z","status":"skipped","version":"0.1.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T21:33:23.493480Z","status":"skipped","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T21:21:57.393285Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"cited_work_retraction","ran_at":"2026-05-19T21:21:56.800361Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"ec2bedbf4b0fdcaf4d9a5d25867c28d27d612c259da2a629dbd71eaa8339262d"},"references":{"count":46,"sample":[{"doi":"","year":2021,"title":"Proceedings of the 38th International Conference on Machine Learning , pages =","work_id":"3596e22d-cb57-4c81-8b7f-b2ba689a57e3","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Proceedings of the 38th International Conference on Machine Learning , pages =","work_id":"02594f6a-5b7e-4ad3-b205-766081f7a7b9","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1007/978-3-319-10602-1_48","year":2014,"title":"Lawrence Zitnick","work_id":"9a41dfbd-d7c8-4b32-943f-59c08fdd6db7","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2014,"title":"Transactions of the Association for Computational Linguistics , volume =","work_id":"123caac8-bd2c-41af-b7c7-a566aa0aada3","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"2022 , url =","work_id":"00bd439c-9544-4b44-9fd0-93111b74b31b","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":46,"snapshot_sha256":"4a4cfdcb5fec68d432b54a84530057c813ab51c0dbd48e8c501ce3d0e79ad8a6","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"d1acef01f1b77ffcb4d5bfb0566240ea0092c781b9c7e1014835019596dede1b"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}