{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:M3EGOZNBQBAPQBOSIUXCNCJUQE","short_pith_number":"pith:M3EGOZNB","schema_version":"1.0","canonical_sha256":"66c86765a18040f805d2452e268934812cbb18a6a1ef6bfd947e79b877d7bd9d","source":{"kind":"arxiv","id":"2605.30257","version":1},"attestation_state":"computed","paper":{"title":"Stable-Layers: Fine-Tuning Image Layer Decomposition Models with VLM-Scored Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Ciara Rowles, Mark Boss, Nikhil Pinnaparaju, Reshinth Adithyan, Vikram Voleti","submitted_at":"2026-05-28T17:20:31Z","abstract_excerpt":"We present Stable-Layers, a reinforcement learning framework that eliminates the need for paired supervision by fine-tuning a pretrained layer decomposition model using only feedback from a vision-language model (VLM). Starting from Qwen-Image-Layered, we apply Flow-GRPO with LoRA adaptation, sampling multiple candidate decompositions per image, scoring them with a VLM, and optimising the policy from group-relative advantages. The key challenge lies in designing a reliable reward signal: VLMs scoring samples in isolation tend to compress their judgements into a narrow band, leaving GRPO with l"},"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":"2605.30257","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-28T17:20:31Z","cross_cats_sorted":[],"title_canon_sha256":"44fdf5b9fac4338896070680fe9dd5962e0514ed0bc67628dce11e0360ffc1f2","abstract_canon_sha256":"865c80228ecfdf21a32311afcd873fb150bdff1f05948c4f7ff3f04f10c64e50"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T02:06:14.242099Z","signature_b64":"PzjAJ6k2xplS02xi1uq58pNi+e9wGSHw7zJJ5gRN8vSa6KT7pdRYGT43ti7VzAu/CyZ+WYU9mBXosK7WA3h6Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"66c86765a18040f805d2452e268934812cbb18a6a1ef6bfd947e79b877d7bd9d","last_reissued_at":"2026-05-29T02:06:14.241674Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T02:06:14.241674Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stable-Layers: Fine-Tuning Image Layer Decomposition Models with VLM-Scored Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Ciara Rowles, Mark Boss, Nikhil Pinnaparaju, Reshinth Adithyan, Vikram Voleti","submitted_at":"2026-05-28T17:20:31Z","abstract_excerpt":"We present Stable-Layers, a reinforcement learning framework that eliminates the need for paired supervision by fine-tuning a pretrained layer decomposition model using only feedback from a vision-language model (VLM). Starting from Qwen-Image-Layered, we apply Flow-GRPO with LoRA adaptation, sampling multiple candidate decompositions per image, scoring them with a VLM, and optimising the policy from group-relative advantages. The key challenge lies in designing a reliable reward signal: VLMs scoring samples in isolation tend to compress their judgements into a narrow band, leaving GRPO with l"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.30257","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/2605.30257/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":"2605.30257","created_at":"2026-05-29T02:06:14.241742+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.30257v1","created_at":"2026-05-29T02:06:14.241742+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.30257","created_at":"2026-05-29T02:06:14.241742+00:00"},{"alias_kind":"pith_short_12","alias_value":"M3EGOZNBQBAP","created_at":"2026-05-29T02:06:14.241742+00:00"},{"alias_kind":"pith_short_16","alias_value":"M3EGOZNBQBAPQBOS","created_at":"2026-05-29T02:06:14.241742+00:00"},{"alias_kind":"pith_short_8","alias_value":"M3EGOZNB","created_at":"2026-05-29T02:06:14.241742+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/M3EGOZNBQBAPQBOSIUXCNCJUQE","json":"https://pith.science/pith/M3EGOZNBQBAPQBOSIUXCNCJUQE.json","graph_json":"https://pith.science/api/pith-number/M3EGOZNBQBAPQBOSIUXCNCJUQE/graph.json","events_json":"https://pith.science/api/pith-number/M3EGOZNBQBAPQBOSIUXCNCJUQE/events.json","paper":"https://pith.science/paper/M3EGOZNB"},"agent_actions":{"view_html":"https://pith.science/pith/M3EGOZNBQBAPQBOSIUXCNCJUQE","download_json":"https://pith.science/pith/M3EGOZNBQBAPQBOSIUXCNCJUQE.json","view_paper":"https://pith.science/paper/M3EGOZNB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.30257&json=true","fetch_graph":"https://pith.science/api/pith-number/M3EGOZNBQBAPQBOSIUXCNCJUQE/graph.json","fetch_events":"https://pith.science/api/pith-number/M3EGOZNBQBAPQBOSIUXCNCJUQE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/M3EGOZNBQBAPQBOSIUXCNCJUQE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/M3EGOZNBQBAPQBOSIUXCNCJUQE/action/storage_attestation","attest_author":"https://pith.science/pith/M3EGOZNBQBAPQBOSIUXCNCJUQE/action/author_attestation","sign_citation":"https://pith.science/pith/M3EGOZNBQBAPQBOSIUXCNCJUQE/action/citation_signature","submit_replication":"https://pith.science/pith/M3EGOZNBQBAPQBOSIUXCNCJUQE/action/replication_record"}},"created_at":"2026-05-29T02:06:14.241742+00:00","updated_at":"2026-05-29T02:06:14.241742+00:00"}