{"paper":{"title":"REVISOR: Beyond Textual Reflection, Towards Multimodal Introspective Reasoning in Long-Form Video Understanding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"REVISOR lets multimodal models reflect on both text and specific video segments to improve long-form video reasoning.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Boshen Xu, Hao Yin, Jian Luan, Jianzhong Ju, Jiaze Li, Jingyang Chen, Wenhui Tan, Yijing Chen, Yuxun Qu, Zhenbo Luo","submitted_at":"2025-11-17T06:25:12Z","abstract_excerpt":"Self-reflection mechanisms that rely on purely text-based rethinking processes perform well in most multimodal tasks. However, when directly applied to long-form video understanding scenarios, they exhibit clear limitations. The fundamental reasons for this lie in two points: (1)long-form video understanding involves richer and more dynamic visual input, meaning rethinking only the text information is insufficient and necessitates a further rethinking process specifically targeting visual information; (2) purely text-based reflection mechanisms lack cross-modal interaction capabilities, preven"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"REVISOR enables MLLMs to collaboratively construct introspective reflection processes across textual and visual modalities, significantly enhancing their reasoning capability for long-form video understanding without requiring supplementary supervised fine-tuning or external models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That adding visual segment rethinking and cross-modal interaction during reflection will overcome the stated limitations of purely text-based reflection when applied to long-form video, and that the DADR reward will produce genuine causal alignment rather than spurious correlations.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"REVISOR adds multimodal visual-text reflection and a Dual Attribution Decoupled Reward to improve long-form video reasoning in MLLMs without extra supervised fine-tuning.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"REVISOR lets multimodal models reflect on both text and specific video segments to improve long-form video reasoning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"91a18ec122b4db5b88fc65cebf7e8798e2038cdc325cce04da0abd343558da76"},"source":{"id":"2511.13026","kind":"arxiv","version":3},"verdict":{"id":"aa920106-99ce-4736-a57b-b65771deff79","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T22:16:26.945472Z","strongest_claim":"REVISOR enables MLLMs to collaboratively construct introspective reflection processes across textual and visual modalities, significantly enhancing their reasoning capability for long-form video understanding without requiring supplementary supervised fine-tuning or external models.","one_line_summary":"REVISOR adds multimodal visual-text reflection and a Dual Attribution Decoupled Reward to improve long-form video reasoning in MLLMs without extra supervised fine-tuning.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That adding visual segment rethinking and cross-modal interaction during reflection will overcome the stated limitations of purely text-based reflection when applied to long-form video, and that the DADR reward will produce genuine causal alignment rather than spurious correlations.","pith_extraction_headline":"REVISOR lets multimodal models reflect on both text and specific video segments to improve long-form video reasoning."},"references":{"count":66,"sample":[{"doi":"","year":2025,"title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning","work_id":"40b60d06-dc1c-4799-b75d-ff1eca653049","ref_index":1,"cited_arxiv_id":"2507.19457","is_internal_anchor":true},{"doi":"","year":2025,"title":"Qwen2.5-VL Technical Report","work_id":"69dffacb-bfe8-442d-be86-48624c60426f","ref_index":2,"cited_arxiv_id":"2502.13923","is_internal_anchor":true},{"doi":"","year":null,"title":"arXiv preprint arXiv:2412.12075 , year=","work_id":"011136c4-f8a3-4d2a-a10f-3e6548f6353f","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Rextime: A benchmark suite for reasoning-across-time in videos.Advances in Neural In- formation Processing Systems, 37:28662–28673, 2024","work_id":"c089f56c-97f6-4d1c-8298-aa2951cd6b1c","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Sharegpt4video: Improving video understand- ing and generation with better captions.Advances in Neural Information Processing Systems, 37:19472–19495, 2024","work_id":"e2ac25e4-27d8-44d5-8a11-c4a0da16b56c","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":66,"snapshot_sha256":"d08bee5f79ec35990e72941b6693e4f652b6e8ba0e5d84c2090eb03533e8e92d","internal_anchors":21},"formal_canon":{"evidence_count":2,"snapshot_sha256":"8df5d36fdb5e71acfd9a9e2c8eba8b9182f286f7ea7a4b9d14d11bd149f7e633"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}