mlsanityData to 13 Oct 2026

Look-Back: Implicit Visual Re-focusing in MLLM Reasoning

arXiv 2507.03019 ↗First posted 2 Jul 2025cs.CV cs.LGcited by 55

On the map: LLM reasoning → Multimodal reasoning

also in 1 more
Abstract (arXiv)

Multimodal Large Language Models (MLLMs) have achieved remarkable progress in multimodal reasoning. However, they often excessively rely on textual information during the later stages of inference, neglecting the crucial integration of visual input. Current methods typically address this by explicitly injecting visual information to guide the reasoning process. In this work, through an analysis of MLLM attention patterns, we made an intriguing observation: with appropriate guidance, MLLMs can spontaneously re-focus their attention on visual inputs during the later stages of reasoning, even without explicit visual information injection. This spontaneous shift in focus suggests that MLLMs are intrinsically capable of performing visual fusion reasoning. Building on this insight, we introduce Look-Back, an implicit approach designed to guide MLLMs to ``look back" at visual information in a self-directed manner during reasoning. Look-Back empowers the model to autonomously determine when, where, and how to re-focus on visual inputs, eliminating the need for explicit model-structure constraints or additional input. We demonstrate that Look-Back significantly enhances the model's reasoning and perception capabilities, as evidenced by extensive empirical evaluations on multiple multimodal benchmarks.

Key facts. Cited by 55 papers in the map (data through 13 Oct 2026). Placed in Multimodal reasoning, one of 17 directions in LLM reasoning. Builds on 84 of its 110 references that are in the map. First posted on arXiv 2 Jul 2025.

Builds on

84 of 110 references.

Closest ancestors · 42

Same direction, newest first

  1. 2025VGR: Visual Grounded Reasoning13 Jun
  2. 2025SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning2 Jun
  3. 2025Grounded Reinforcement Learning for Visual Reasoning29 May
  4. 2025Infi-MMR: Curriculum-based Unlocking Multimodal Reasoning via Phased Reinforcement Learning in Multimodal Small Language Models29 May
  5. 2025Thinking with Generated Images28 May
  6. 2025v1: Learning to Point Visual Tokens for Multimodal Grounded Reasoning24 May
  7. 2025R1-ShareVL: Incentivizing Reasoning Capability of Multimodal Large Language Models via Share-GRPO22 May
  8. 2025GRIT: Teaching MLLMs to Think with Images21 May
  9. 2025Adaptive Chain-of-Focus Reasoning via Dynamic Visual Search and Zooming for Efficient VLMs21 May
  10. 2025Visionary-R1: Mitigating Shortcuts in Visual Reasoning with Reinforcement Learning20 May
  11. 2025DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning20 May
  12. 2025OpenThinkIMG: Learning to Think with Images via Visual Tool Reinforcement Learning13 May
30 more
  1. 2025NoisyRollout: Reinforcing Visual Reasoning with Data Augmentation17 Apr
  2. 2025VisuoThink: Empowering LVLM Reasoning with Multimodal Tree Search12 Apr
  3. 2025SFT or RL? An Early Investigation into Training R1-Like Reasoning Large Vision-Language Models10 Apr
  4. 2025VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning10 Apr
  5. 2025Perception-R1: Pioneering Perception Policy with Reinforcement Learning10 Apr
  6. 2025SoTA with Less: MCTS-Guided Sample Selection for Data-Efficient Visual Reasoning Self-Improvement10 Apr
  7. 2025Video-R1: Reinforcing Video Reasoning in MLLMs27 Mar
  8. 2025Reason-RFT: Reinforcement Fine-Tuning for Visual Reasoning of Vision Language Models26 Mar
  9. 2025OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles21 Mar
  10. 2025Mitigating Visual Forgetting via Take-along Visual Conditioning for Multi-modal Long CoT Reasoning17 Mar
  11. 2025Multimodal Chain-of-Thought Reasoning: A Comprehensive Survey16 Mar
  12. 2025R1-Onevision: Advancing Generalized Multimodal Reasoning through Cross-Modal Formalization13 Mar
  13. 2025LMM-R1: Empowering 3B LMMs with Strong Reasoning Abilities Through Two-Stage Rule-Based RL10 Mar
  14. 2025MM-Eureka: Exploring the Frontiers of Multimodal Reasoning with Rule-based Reinforcement Learning10 Mar
  15. 2025Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models9 Mar
  16. 2025Seg-Zero: Reasoning-Chain Guided Segmentation via Cognitive Reinforcement9 Mar
  17. 2025R1-Zero's "Aha Moment" in Visual Reasoning on a 2B Non-SFT Model7 Mar
  18. 2025Visual-RFT: Visual Reinforcement Fine-Tuning3 Mar
  19. 2025Imagine while Reasoning in Space: Multimodal Visualization-of-Thought13 Jan
  20. 2025LlamaV-o1: Rethinking Step-by-step Visual Reasoning in LLMs10 Jan
  21. 2025ReFocus: Visual Editing as a Chain of Thought for Structured Image Understanding9 Jan
  22. 2025Virgo: A Preliminary Exploration on Reproducing o1-like MLLM3 Jan
  23. 2024LATTE: Learning to Think with Vision Specialists7 Dec
  24. 2024MAmmoTH-VL: Eliciting Multimodal Reasoning with Instruction Tuning at Scale6 Dec
  25. 2024LLaVA-CoT: Let Vision Language Models Reason Step-by-Step15 Nov
  26. 2024We-Math: Does Your Large Multimodal Model Achieve Human-like Mathematical Reasoning?1 Jul
  27. 2024Math-LLaVA: Bootstrapping Mathematical Reasoning for Multimodal Large Language Models25 Jun
  28. 2024Visual Sketchpad: Sketching as a Visual Chain of Thought for Multimodal Language Models13 Jun
  29. 2023MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts3 Oct
  30. 2023Multimodal Chain-of-Thought Reasoning in Language Models2 Feb

Other references · 42

  1. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,091
  2. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,247
  3. 2024GPT-4o System Cardcited by 4,536
  4. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 4,191
  5. 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,729
  6. 2023Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyondcited by 2,204
  7. 2024HybridFlow: A Flexible and Efficient RLHF Frameworkcited by 2,003
  8. 2025InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Modelscited by 1,798
  9. 2023MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Modelscited by 1,637
  10. 2025s1: Simple test-time scalingcited by 1,322
  11. 2024OpenAI o1 System Cardcited by 1,233
  12. 2025Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?cited by 964
30 more
  1. 2021Inter-GPS: Interpretable Geometry Problem Solving with Formal Language and Symbolic Reasoningcited by 378
  2. 2025R1-VL: Learning to Reason with Multimodal Large Language Models via Step-wise Group Relative Policy Optimizationcited by 299
  3. 2024HuatuoGPT-o1, Towards Medical Complex Reasoning with LLMscited by 260
  4. 2023G-LLaVA: Solving Geometric Problem with Multi-Modal Large Language Modelcited by 210
  5. 2025MLLMs Know Where to Look: Training-free Perception of Small Visual Details with Multimodal LLMscited by 173
  6. 2025FutureSightDrive: Thinking Visually with Spatio-Temporal CoT for Autonomous Drivingcited by 166
  7. 2019Neural-Symbolic Computing: An Effective Methodology for Principled Integration of Machine Learning and Reasoningcited by 150
  8. 2025Reinforcing Spatial Reasoning in Vision-Language Models with Interwoven Thinking and Visual Drawingcited by 132
  9. 2025LLM Post-Training: A Deep Dive into Reasoning Large Language Modelscited by 131
  10. 2024ZoomEye: Enhancing Multimodal LLMs with Human-Like Zooming Capabilities through Tree-Based Image Explorationcited by 101
  11. 2025Visual Planning: Let's Think Only with Imagescited by 85
  12. 2024Look Twice Before You Answer: Memory-Space Visual Retracing for Hallucination Mitigation in Multimodal Large Language Modelscited by 72
  13. 2025UniWorld-V1: High-Resolution Semantic Encoders for Unified Visual Understanding and Generationcited by 65
  14. 2025Embodied-Reasoner: Synergizing Visual Search, Reasoning, and Action for Embodied Interactive Taskscited by 43
  15. 2025Reflective Planning: Vision-Language Models for Multi-Stage Long-Horizon Robotic Manipulationcited by 41
  16. 2025PhyX: Does Your Model Have the "Wits" for Physical Reasoning?cited by 38
  17. 2025Embodied-R: Collaborative Framework for Activating Embodied Spatial Reasoning in Foundation Models via Reinforcement Learningcited by 37
  18. 2025Jigsaw-R1: A Study of Rule-based Visual Reinforcement Learning with Jigsaw Puzzlescited by 36
  19. 2025MEDA: Dynamic KV Cache Allocation for Efficient Multimodal Long-Context Inferencecited by 31
  20. 2025X-Reasoner: Towards Generalizable Reasoning Across Modalities and Domainscited by 31
  21. 2025One RL to See Them All: Visual Triple Unified Reinforcement Learningcited by 30
  22. 2024Next Patch Prediction for Autoregressive Visual Generationcited by 25
  23. 2025Attention Reallocation: Towards Zero-cost and Controllable Hallucination Mitigation of MLLMscited by 23
  24. 2025STAR-R1: Spatial TrAnsformation Reasoning by Reinforcing Multimodal LLMscited by 21
  25. 2025AsFT: Anchoring Safety During LLM Fine-Tuning Within Narrow Safety Basincited by 19
  26. 2024Is Parameter Collision Hindering Continual Learning in LLMs?cited by 15
  27. 2025CrowdVLM-R1: Expanding R1 Ability to Vision Language Model for Crowd Counting using Fuzzy Group Relative Policy Rewardcited by 15
  28. 2025LanP: Rethinking the Impact of Language Priors in Large Vision-Language Modelscited by 14
  29. 2025DIP-R1: Deep Inspection and Perception with RL Looking Through and Understanding Complex Scenescited by 11
  30. 2025LangBridge: Interpreting Image as a Combination of Language Embeddingscited by 9

Built on by

Cited by year (2026 to 13 Oct)

Citing papers per year: 2025: 9; 2026: 46.

55 papers.

Sort:

By area:

  1. 2026Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perceptioncited by 48
  2. 2025From Perception to Cognition: A Survey of Vision-Language Interactive Reasoning in Multimodal Large Language Modelscited by 26
  3. 2025More Thought, Less Accuracy? On the Dual Nature of Reasoning in Vision-Language Modelscited by 26
  4. 2025A Survey on Agentic Multimodal Large Language Modelscited by 22
  5. 2025Perceptual-Evidence Anchored Reinforced Learning for Multimodal Reasoningcited by 20
  6. 2025REVISOR: Beyond Textual Reflection, Towards Multimodal Introspective Reasoning in Long-Form Video Understandingcited by 15
  7. 2025Reinforcement Learning with Inverse Rewards for World Model Post-trainingcited by 12
  8. 2026PatchCue: Enhancing Vision-Language Model Reasoning with Patch-Based Visual Cuescited by 8
  9. 2026Analyzing Reasoning Consistency in Large Multimodal Models under Cross-Modal Conflictscited by 7
  10. 2026H-OPD: Confidence Aware Heterogeneous Multi-Teacher Multimodal On-policy Distillationcited by 7
  11. 2025Segmentation as A Plug-and-Play Capability for Frozen Multimodal LLMscited by 5
  12. 2025Learning When to Look: A Disentangled Curriculum for Strategic Perception in Multimodal Reasoningcited by 5
38 more
  1. 2026DVGBench: Implicit-to-Explicit Visual Grounding Benchmark in UAV Imagery with Large Vision-Language Modelscited by 4
  2. 2026Do MLLMs Really See It: Reinforcing Visual Attention in Multimodal LLMscited by 4
  3. 2026CARE: Towards Clinical Accountability in Multi-Modal Medical Reasoning with an Evidence-Grounded Agentic Frameworkcited by 4
  4. 2026Visual Reasoning through Tool-supervised Reinforcement Learningcited by 4
  5. 2026Dr. Seg: Revisiting GRPO Training for Visual Large Language Models through Perception-Oriented Designcited by 3
  6. 2026When Thinking Hurts: Mitigating Visual Forgetting in Video Reasoning via Frame Repetitioncited by 3
  7. 2026Visual Enhanced Depth Scaling for Multimodal Latent Reasoningcited by 3
  8. 2026Thinking With Tools, Not With Pixels: Tool Calls as Text Scaffolds for Visual Reasoningcited by 3
  9. 2026CSR-Bench: A Benchmark for Evaluating the Cross-modal Safety and Reliability of MLLMscited by 2
  10. 2026Bridging Modality Disconnect in Self-Reflection via Closed-Loop Visually Grounded Verificationcited by 2
  11. 2026Understanding and Mitigating Hallucinations in Multimodal Chain-of-Thought Modelscited by 2
  12. 2026From Failure to Feedback: Group Revision Unlocks Hard Cases in Object-Level Groundingcited by 2
  13. 2026One-Way Policy Optimization for Self-Evolving LLMscited by 2
  14. 2025ChainV: Atomic Visual Hints Make Multimodal Reasoning Shorter and Bettercited by 1
  15. 2026CPPO: Contrastive Perception Policy Optimization for VLM Agentscited by 1
  16. 2026When RAG Hurts: Diagnosing and Mitigating Attention Distraction in Retrieval-Augmented LVLMscited by 1
  17. 2026Canvas-of-Thought: Grounding Reasoning via Mutable Structured Statescited by 1
  18. 2026VisRef: Visual Refocusing while Thinking Improves Test-Time Scaling in Multi-Modal Large Reasoning Modelscited by 1
  19. 2026All Roads Lead to Rome: Incentivizing Divergent Thinking in Vision-Language Modelscited by 1
  20. 2026Spotlight and Shadow: Attention-Guided Dual-Anchor Introspective Decoding for MLLM Hallucination Mitigationcited by 1
  21. 2026RAVE: Re-Allocating Visual Attention in Large Multimodal Modelscited by 1
  22. 2026VisionPulse: Dynamic Visual Sparsity for Efficient Multimodal Reasoningcited by 1
  23. 2026Teaching the Way, Not the Answer: Privileged Tutoring Distillation for Multimodal Policy Optimizationcited by 1
  24. 2026DyCo-RL: Dynamic Cross-Modal Coordination for Visual Reasoningcited by 1
  25. 2026LASER: A Corrective Lens for LVLMs via Visual Attention Preservation and Sink Suppressioncited by 1
  26. 2026Scene Graph Thinking: Reinforcing Structured Visual Reasoning for Multimodal Large Language Modelscited by 1
  27. 2026ReGround: Restoring Visual Grounding in Multi-Step Reasoning through Self-Diagnosis and Visual Re-Examinationcited by 1
  28. 2026Temporal Self-Distillation: Learning Visual State Tracking in Videos Without Supervisioncited by 1
  29. 2026Mimic Human Cognition, Master Multi-Image Reasoning: A Meta-Action Framework for Enhanced Visual Understandingcited by 0
  30. 2026CoTZero: Annotation-Free Human-Like Vision Reasoning via Hierarchical Synthetic CoTcited by 0
  31. 2026Reflect to Inform: Boosting Multimodal Reasoning via Information-Gain-Driven Verificationcited by 0
  32. 2026Reasoning Portability: Guiding Continual Learning for MLLMs in the RLVR Eracited by 0
  33. 2026AgroTools: A Benchmark for Tool-Augmented Multimodal Agents in Agriculturecited by 0
  34. 2026Clipping Bottleneck: Stabilizing RLVR via Stochastic Recovery of Near-Boundary Signalscited by 0
  35. 2026LOCUS: Local Visual Cue Search for Enhancing Fine-Grained Perception in Multimodal Large Language Modelscited by 0
  36. 2026From Structure to Synergy: A Survey of Vision-Language Perception Paradigm Evolution in Multimodal Large Language Modelscited by 0
  37. 2026Visually Grounded Self-Reflection for Vision-Language Models via Reinforcement Learningcited by 0
  38. 2026SpatialAfford: Teaching Compact VLMs Where to Look and Where to Ground for Affordancecited by 0

50 most cited of 55 All 55 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026MM-OPD: Towards One More Bottleneck Between Perception and Reasoning
  2. 2026Where Does Vision Meet Language? Understanding and Refining Visual Fusion in MLLMs via Contrastive Attention
  3. 2026LookWise: Knowing When and Where to Look for Fine-Grained Visual Reasoning in Multimodal Large Language Models
  4. 2026SwimBird: Eliciting Switchable Reasoning Mode in Hybrid Autoregressive MLLMs
  5. 2026The Ebb and Flow of Multimodal Focus: Scheduling Visual Relay Windows for Grounded VLM Reasoning
  6. 2025See, Think, Learn: A Self-Taught Multimodal Reasoner
  7. 2026Balancing Efficiency and Efficacy: Training-Free Attention-Guided Switching Between Explicit and Latent Thoughts for MLLMs
  8. 2026Look on Demand: A Cognitive Scheduling Framework for Visual Evidence Acquisition in Multimodal Reasoning