{"paper":{"title":"Edit-GRPO: A Locality-Preserving Policy Optimization Framework for Image Editing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Edit-GRPO decouples editing and preservation objectives with region-specific signals to keep image edits localized.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bo Zheng, Houqiang Li, Litong Gong, Shaodong Xu, Tiezheng Ge, Wengang Zhou, Zexian Li, Zhendong Wang","submitted_at":"2026-05-16T12:05:39Z","abstract_excerpt":"A fundamental challenge in image editing lies in preserving spatial locality: edits should improve targeted content without inadvertently altering surrounding regions. However, most optimization-based editing approaches treat images as holistic entities, causing global policy updates that undermine locality and introduce undesired context changes. We observe that this issue stems from a mismatch between localized editing intent and globally applied optimization signals. Motivated by this insight, we propose Edit-GRPO, preserving Locality while optimizing image editing, a locality-preserving po"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Extensive experiments across diverse image editing scenarios demonstrate that Edit-GRPO significantly improves locality preservation while maintaining strong editing performance compared to existing optimization-based methods.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that decoupling editing and preservation objectives through region-specific optimization signals will align policy updates with the spatial structure of editing tasks without requiring additional mechanisms to handle boundary effects or context interactions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Edit-GRPO decouples editing and preservation objectives via region-specific signals in a policy optimization framework to improve locality in image editing tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Edit-GRPO decouples editing and preservation objectives with region-specific signals to keep image edits localized.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9369d5179d0c4dd6f66aadb770442d3ed6f6c83d3ac9b695a8cda482173be52c"},"source":{"id":"2605.16951","kind":"arxiv","version":1},"verdict":{"id":"c45acb61-ca65-4f63-b876-005172fc8695","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:41:04.325420Z","strongest_claim":"Extensive experiments across diverse image editing scenarios demonstrate that Edit-GRPO significantly improves locality preservation while maintaining strong editing performance compared to existing optimization-based methods.","one_line_summary":"Edit-GRPO decouples editing and preservation objectives via region-specific signals in a policy optimization framework to improve locality in image editing tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that decoupling editing and preservation objectives through region-specific optimization signals will align policy updates with the spatial structure of editing tasks without requiring additional mechanisms to handle boundary effects or context interactions.","pith_extraction_headline":"Edit-GRPO decouples editing and preservation objectives with region-specific signals to keep image edits localized."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16951/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.101862Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T20:50:51.181185Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"cited_work_retraction","ran_at":"2026-05-19T19:52:11.122595Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T18:41:56.238257Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.322016Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"26fc7a8a11745505fb2a31ecaf5e1879da5b22e6545c307c1ad96cb3278ab121"},"references":{"count":48,"sample":[{"doi":"","year":2024,"title":"Humanedit: A high-quality human-rewarded dataset for instruction-based image editing","work_id":"9924b111-50c3-43ab-862e-d948bcc1fb17","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Qwen3-VL Technical Report","work_id":"1fe243aa-e3c0-4da6-b391-4cbcfc88d5c0","ref_index":2,"cited_arxiv_id":"2511.21631","is_internal_anchor":true},{"doi":"","year":2023,"title":"Training Diffusion Models with Reinforcement Learning","work_id":"67684dda-3930-452a-b91a-36cbb8e2e219","ref_index":3,"cited_arxiv_id":"2305.13301","is_internal_anchor":true},{"doi":"","year":2022,"title":"In- structpix2pix: Learning to follow image editing instructions","work_id":"de9c4a95-36ff-4e97-a0c0-3cca39f60b1a","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"HiDream-I1: A High-Efficient Image Generative Foundation Model with Sparse Diffusion Transformer","work_id":"68d4c0f7-3dfd-438d-a823-6a93fd0a835d","ref_index":5,"cited_arxiv_id":"2505.22705","is_internal_anchor":true}],"resolved_work":48,"snapshot_sha256":"c1c30ed6479336a41d5de550fcc3aff55b27df71269967d2fd5f39081f07c86e","internal_anchors":25},"formal_canon":{"evidence_count":2,"snapshot_sha256":"dc0af3956b4d46f69dead2c11dec649ae76dc8a420817883207e7685cfd8b793"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}