mlsanityData to 7 Oct 2026

MG-Thinker: Bi-Axial Self-Reflection for Multi-Image Reasoning Grounding

arXiv 2609.37374 ↗First posted 29 Sep 2026cs.CV cs.MMcited by 0

On the map: LLM reasoning → Multimodal reasoning

Abstract (arXiv)

Reinforcement learning (RL) has recently delivered substantial gains in multimodal reasoning, opening a promising route for fine-grained visual perception. Yet for multi-image reasoning grounding (MRG), reasoning over real-world multi-image contexts toward pixel-precise localization, existing RL-based approaches overlook two characteristics intrinsic to this paradigm: a coarse-to-fine hierarchical reasoning pattern, and heterogeneously distributed task--sample difficulties. In this work, we present MG-Thinker, a post-training RL framework that advances a new MRG paradigm featuring such hierarchical reasoning, supported by a curated 25K MRG dataset with task-adaptive Chain-of-Thought (CoT) annotations that elicit multi-perspective evidence before conclusion. To remedy the heterogeneous task--sample difficulties, we further propose Bi-Axial DAPO (BiA-DAPO), which decomposes rollout advantages along an intra-group signal axis and an inter-group competence axis through two complementary mechanisms, both grounded on our defined candidate pool for stable group-level statistics. Extensive experiments show that MG-Thinker achieves state-of-the-art performance on multi-image reasoning grounding while consistently improving generalization across multi-image understanding and diverse multimodal benchmarks.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Multimodal reasoning, one of 17 directions in LLM reasoning. Builds on 27 of its 45 references that are in the map. First posted on arXiv 29 Sep 2026.

Builds on

27 of 45 references.

Closest ancestors · 6

Same direction, newest first

  1. 2025GRIT: Teaching MLLMs to Think with Images21 May
  2. 2025UniVG-R1: Reasoning Guided Universal Visual Grounding with Reinforcement Learning20 May
  3. 2025VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning10 Apr
  4. 2024LLaVA-CoT: Let Vision Language Models Reason Step-by-Step15 Nov
  5. 2024One Token to Seg Them All: Language Instructed Reasoning Segmentation in Videos29 Sep
  6. 2024Visual Sketchpad: Sketching as a Visual Chain of Thought for Multimodal Language Models13 Jun

Other references · 21

  1. 2025Qwen2.5-VL Technical Reportcited by 5,418
  2. 2023Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detectioncited by 3,274
  3. 2023InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Taskscited by 2,534
  4. 2025InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Modelscited by 1,769
  5. 2023Kosmos-2: Grounding Multimodal Large Language Models to the Worldcited by 1,075
  6. 2024MiniCPM-V: A GPT-4V Level MLLM on Your Phonecited by 1,061
  7. 2016Modeling Context in Referring Expressionscited by 1,054
  8. 2015Generation and Comprehension of Unambiguous Object Descriptionscited by 897
  9. 2023Shikra: Unleashing Multimodal LLM's Referential Dialogue Magiccited by 891
  10. 2024Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMscited by 822
  11. 2024BLINK: Multimodal Large Language Models Can See but Not Perceivecited by 529
  12. 2023Ferret: Refer and Ground Anything Anywhere at Any Granularitycited by 498
9 more
  1. 2023GLaMM: Pixel Grounding Large Multimodal Modelcited by 447
  2. 2023GPT4RoI: Instruction Tuning Large Language Model on Region-of-Interestcited by 300
  3. 2024MANTIS: Interleaved Multi-Image Instruction Tuningcited by 212
  4. 2024MuirBench: A Comprehensive Benchmark for Robust Multi-image Understandingcited by 178
  5. 2024VisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Taskscited by 136
  6. 2025Efficient Reinforcement Finetuning via Adaptive Curriculum Learningcited by 88
  7. 2024VoCoT: Unleashing Visually Grounded Multi-Step Reasoning in Large Multi-Modal Modelscited by 42
  8. 2024MIBench: Evaluating Multimodal Large Language Models over Multiple Imagescited by 36
  9. 2024Griffon v2: Advancing Multimodal Perception with High-Resolution Scaling and Visual-Language Co-Referringcited by 29

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Ground-R1: Incentivizing Grounded Visual Reasoning via Reinforcement Learning
  2. 2025V-Thinker: Interactive Thinking with Images
  3. 2025MIRG-RL: Multi-Image Reasoning and Grounding with Reinforcement Learning
  4. 2026Faithful GRPO: Improving Visual Spatial Reasoning in Multimodal Language Models via Constrained Policy Optimization
  5. 2025Guiding the Inner Eye: A Framework for Hierarchical and Flexible Visual Grounded Reasoning
  6. 2026See It, Say It, Sorted: An Iterative Training-Free Framework for Visually-Grounded Multimodal Reasoning in LVLMs
  7. 2026GeM-VG: Towards Generalized Multi-image Visual Grounding with Multimodal Large Language Models
  8. 2025More Thought, Less Accuracy? On the Dual Nature of Reasoning in Vision-Language Models