{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:YPKS2S7Z3HU6CDZU5UKWPYLUUS","short_pith_number":"pith:YPKS2S7Z","schema_version":"1.0","canonical_sha256":"c3d52d4bf9d9e9e10f34ed1567e174a4ad07b39069d10c03e3c222556c05034d","source":{"kind":"arxiv","id":"2512.00814","version":3},"attestation_state":"computed","paper":{"title":"IRPO: Boosting Image Restoration via Post-training GRPO","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Boyuan Jiang, Donghao Luo, Haoang Li, Haoxuan Xu, Jinlong Peng, Ruolin Shen, Shuicheng Yan, Tianfu Li, Xiaobin Hu, Yi Liu","submitted_at":"2025-11-30T09:42:24Z","abstract_excerpt":"Post-training has become effective for high-level generation, but its role in low-level vision remains underexplored. Existing image restoration methods often rely on fixed pixel-wise fitting to ground-truth images, which can lead to over-smoothing and weak generalization. We propose IRPO, a GRPO-based post-training framework for deterministic restoration models. IRPO is built around two axes: data formulation and reward modeling. For data formulation, we select the 30% underperforming samples from the pre-training stage, which improves both accuracy and training efficiency. For reward modelin"},"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":"2512.00814","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-11-30T09:42:24Z","cross_cats_sorted":[],"title_canon_sha256":"4d92faefcc1a71df1a1535767afe0281906d2c0ac705d634d0d97f5d6eac8fcd","abstract_canon_sha256":"cfc94517ff3394a67be18674ce90972bff534d55fb3b1a62f37b171f28ecd115"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:33.671961Z","signature_b64":"AqjG9nv9YSfcdI0c+r2oLXxQ/4jqbBnIDTXylN3MYHDEFhlA8g0VxL2xmlsk4cRlNWKa/A9N8pHIWUPePn4XBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c3d52d4bf9d9e9e10f34ed1567e174a4ad07b39069d10c03e3c222556c05034d","last_reissued_at":"2026-05-28T01:04:33.671452Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:33.671452Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"IRPO: Boosting Image Restoration via Post-training GRPO","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Boyuan Jiang, Donghao Luo, Haoang Li, Haoxuan Xu, Jinlong Peng, Ruolin Shen, Shuicheng Yan, Tianfu Li, Xiaobin Hu, Yi Liu","submitted_at":"2025-11-30T09:42:24Z","abstract_excerpt":"Post-training has become effective for high-level generation, but its role in low-level vision remains underexplored. Existing image restoration methods often rely on fixed pixel-wise fitting to ground-truth images, which can lead to over-smoothing and weak generalization. We propose IRPO, a GRPO-based post-training framework for deterministic restoration models. IRPO is built around two axes: data formulation and reward modeling. For data formulation, we select the 30% underperforming samples from the pre-training stage, which improves both accuracy and training efficiency. For reward modelin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2512.00814","kind":"arxiv","version":3},"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/2512.00814/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":"2512.00814","created_at":"2026-05-28T01:04:33.671516+00:00"},{"alias_kind":"arxiv_version","alias_value":"2512.00814v3","created_at":"2026-05-28T01:04:33.671516+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2512.00814","created_at":"2026-05-28T01:04:33.671516+00:00"},{"alias_kind":"pith_short_12","alias_value":"YPKS2S7Z3HU6","created_at":"2026-05-28T01:04:33.671516+00:00"},{"alias_kind":"pith_short_16","alias_value":"YPKS2S7Z3HU6CDZU","created_at":"2026-05-28T01:04:33.671516+00:00"},{"alias_kind":"pith_short_8","alias_value":"YPKS2S7Z","created_at":"2026-05-28T01:04:33.671516+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/YPKS2S7Z3HU6CDZU5UKWPYLUUS","json":"https://pith.science/pith/YPKS2S7Z3HU6CDZU5UKWPYLUUS.json","graph_json":"https://pith.science/api/pith-number/YPKS2S7Z3HU6CDZU5UKWPYLUUS/graph.json","events_json":"https://pith.science/api/pith-number/YPKS2S7Z3HU6CDZU5UKWPYLUUS/events.json","paper":"https://pith.science/paper/YPKS2S7Z"},"agent_actions":{"view_html":"https://pith.science/pith/YPKS2S7Z3HU6CDZU5UKWPYLUUS","download_json":"https://pith.science/pith/YPKS2S7Z3HU6CDZU5UKWPYLUUS.json","view_paper":"https://pith.science/paper/YPKS2S7Z","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2512.00814&json=true","fetch_graph":"https://pith.science/api/pith-number/YPKS2S7Z3HU6CDZU5UKWPYLUUS/graph.json","fetch_events":"https://pith.science/api/pith-number/YPKS2S7Z3HU6CDZU5UKWPYLUUS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YPKS2S7Z3HU6CDZU5UKWPYLUUS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YPKS2S7Z3HU6CDZU5UKWPYLUUS/action/storage_attestation","attest_author":"https://pith.science/pith/YPKS2S7Z3HU6CDZU5UKWPYLUUS/action/author_attestation","sign_citation":"https://pith.science/pith/YPKS2S7Z3HU6CDZU5UKWPYLUUS/action/citation_signature","submit_replication":"https://pith.science/pith/YPKS2S7Z3HU6CDZU5UKWPYLUUS/action/replication_record"}},"created_at":"2026-05-28T01:04:33.671516+00:00","updated_at":"2026-05-28T01:04:33.671516+00:00"}