{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:PI53PQ3VFE6H6G4GTX3NH7ANLX","short_pith_number":"pith:PI53PQ3V","schema_version":"1.0","canonical_sha256":"7a3bb7c375293c7f1b869df6d3fc0d5dc7f32ac67fcf303d6eecd6266867bb07","source":{"kind":"arxiv","id":"2511.18719","version":4},"attestation_state":"computed","paper":{"title":"Seeing What Matters: Visual Preference Policy Optimization for Visual Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Chi Zhang, Haibin Huang, Rui Li, Xuelong Li, Yi Zhou, Yuanzhi Liang, Ziqi Ni","submitted_at":"2025-11-24T03:21:17Z","abstract_excerpt":"Reinforcement learning (RL) has become a powerful tool for post-training visual generative models, with Group Relative Policy Optimization (GRPO) increasingly used to align generators with human preferences. However, existing GRPO pipelines rely on a single scalar reward per sample, treating each image or video as a holistic entity and ignoring the rich spatial and temporal structure of visual content. This coarse supervision hinders the correction of localized artifacts and the modeling of fine-grained perceptual cues. We introduce Visual Preference Policy Optimization (ViPO), a GRPO variant "},"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":"2511.18719","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-11-24T03:21:17Z","cross_cats_sorted":[],"title_canon_sha256":"8600f52156e4ac63ec5ef2873ae469f3c1927ae8315e3e1f73051ad85bf82aea","abstract_canon_sha256":"783e60a59b9e1f8461615db0c44c3a12273e463ae28e149c720a14ac3cb4d0a1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:00:28.454924Z","signature_b64":"6zAL4aTMqEAPVrLrRvCz9Hff0B8IjuCH/9q6GqBwXge3GIh+PKgowk6d8hJihaH0rB/armaGxSX3XQCmYd/dDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7a3bb7c375293c7f1b869df6d3fc0d5dc7f32ac67fcf303d6eecd6266867bb07","last_reissued_at":"2026-05-20T00:00:28.454124Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:00:28.454124Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Seeing What Matters: Visual Preference Policy Optimization for Visual Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Chi Zhang, Haibin Huang, Rui Li, Xuelong Li, Yi Zhou, Yuanzhi Liang, Ziqi Ni","submitted_at":"2025-11-24T03:21:17Z","abstract_excerpt":"Reinforcement learning (RL) has become a powerful tool for post-training visual generative models, with Group Relative Policy Optimization (GRPO) increasingly used to align generators with human preferences. However, existing GRPO pipelines rely on a single scalar reward per sample, treating each image or video as a holistic entity and ignoring the rich spatial and temporal structure of visual content. This coarse supervision hinders the correction of localized artifacts and the modeling of fine-grained perceptual cues. We introduce Visual Preference Policy Optimization (ViPO), a GRPO variant "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2511.18719","kind":"arxiv","version":4},"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/2511.18719/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":"2511.18719","created_at":"2026-05-20T00:00:28.454260+00:00"},{"alias_kind":"arxiv_version","alias_value":"2511.18719v4","created_at":"2026-05-20T00:00:28.454260+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2511.18719","created_at":"2026-05-20T00:00:28.454260+00:00"},{"alias_kind":"pith_short_12","alias_value":"PI53PQ3VFE6H","created_at":"2026-05-20T00:00:28.454260+00:00"},{"alias_kind":"pith_short_16","alias_value":"PI53PQ3VFE6H6G4G","created_at":"2026-05-20T00:00:28.454260+00:00"},{"alias_kind":"pith_short_8","alias_value":"PI53PQ3V","created_at":"2026-05-20T00:00:28.454260+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.14274","citing_title":"CreFlow: Corrective Reflow for Sparse-Reward Embodied Video Diffusion RL","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2604.19234","citing_title":"Learning to Credit the Right Steps: Objective-aware Process Optimization for Visual Generation","ref_index":30,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PI53PQ3VFE6H6G4GTX3NH7ANLX","json":"https://pith.science/pith/PI53PQ3VFE6H6G4GTX3NH7ANLX.json","graph_json":"https://pith.science/api/pith-number/PI53PQ3VFE6H6G4GTX3NH7ANLX/graph.json","events_json":"https://pith.science/api/pith-number/PI53PQ3VFE6H6G4GTX3NH7ANLX/events.json","paper":"https://pith.science/paper/PI53PQ3V"},"agent_actions":{"view_html":"https://pith.science/pith/PI53PQ3VFE6H6G4GTX3NH7ANLX","download_json":"https://pith.science/pith/PI53PQ3VFE6H6G4GTX3NH7ANLX.json","view_paper":"https://pith.science/paper/PI53PQ3V","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2511.18719&json=true","fetch_graph":"https://pith.science/api/pith-number/PI53PQ3VFE6H6G4GTX3NH7ANLX/graph.json","fetch_events":"https://pith.science/api/pith-number/PI53PQ3VFE6H6G4GTX3NH7ANLX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PI53PQ3VFE6H6G4GTX3NH7ANLX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PI53PQ3VFE6H6G4GTX3NH7ANLX/action/storage_attestation","attest_author":"https://pith.science/pith/PI53PQ3VFE6H6G4GTX3NH7ANLX/action/author_attestation","sign_citation":"https://pith.science/pith/PI53PQ3VFE6H6G4GTX3NH7ANLX/action/citation_signature","submit_replication":"https://pith.science/pith/PI53PQ3VFE6H6G4GTX3NH7ANLX/action/replication_record"}},"created_at":"2026-05-20T00:00:28.454260+00:00","updated_at":"2026-05-20T00:00:28.454260+00:00"}