mlsanityData to 7 Oct 2026

Progressive-View On-Policy Distillation for Regional-to-Global Transfer in Multimodal LLMs

arXiv 2609.32333 ↗First posted 26 Sep 2026cs.CV cs.AIcited by 0

On the map: Multimodal LLMs / vision-language models → VLM benchmarks and visual reasoning also fits: MLLM training and visual instruction tuning

Abstract (arXiv)

Regional-to-global distillation uses crop-conditioned guidance to improve full-image understanding. The challenge is to effectively transfer the teacher's crop-based advantage to the student's full-image inference. We propose progressive-view on-policy distillation (PVD), which shifts the student's view distribution from the crop toward the full image through an intermediate aspect-preserving padded crop. The padded crop preserves regional content while matching the full image's visual-token grid. Across stages, the view mixture assigns increasing probability to the full image. A lightweight regional-advantage weighting reallocates token-level supervision using the crop-conditioned teacher-student log-probability gap. Evaluated under each sampled input, it applies mild reweighting when the gap is small and emphasizes higher-gap tokens when the gap widens. A Jensen-Shannon metric decomposition interprets this schedule as a shift from matched-input imitation toward the deployment objective. Across benchmarks spanning perception, visual mathematics and general multimodal question answering, PVD-full reaches an average accuracy of 77.51 over three seeds, improving on the reward-free distillation baseline by 2.01 points and on its reward-matched variant by 1.00 point. In the reward-free setting, PVD-distill still gains 1.16 points.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in VLM benchmarks and visual reasoning, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 19 of its 21 references that are in the map. First posted on arXiv 26 Sep 2026.

Builds on

19 of 21 references.

Closest ancestors · 3

Same direction, newest first

  1. 2024Are We on the Right Way for Evaluating Large Vision-Language Models?29 Mar
  2. 2024MathVerse: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?21 Mar
  3. 2023MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts3 Oct

Other references · 16

  1. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  2. 2025Qwen3 Technical Reportcited by 8,195
  3. 2023MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGIcited by 2,044
  4. 2024HybridFlow: A Flexible and Efficient RLHF Frameworkcited by 1,980
  5. 2023On-Policy Distillation of Language Models: Learning from Self-Generated Mistakescited by 840
  6. 2024Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMscited by 822
  7. 2016A Diagram Is Worth A Dozen Imagescited by 793
  8. 2023V*: Guided Visual Search as a Core Mechanism in Multimodal LLMscited by 459
  9. 2015Policy Distillationcited by 414
  10. 2024Divide, Conquer and Combine: A Training-Free Framework for High-Resolution Image Perception in Multimodal Large Language Modelscited by 151
  11. 2024ZoomEye: Enhancing Multimodal LLMs with Human-Like Zooming Capabilities through Tree-Based Image Explorationcited by 99
  12. 2026Vision-OPD: Learning to See Fine Details for Multimodal LLMs via On-Policy Self-Distillationcited by 71
4 more
  1. 2026Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perceptioncited by 47
  2. 2026Not All Disagreement Is Learnable: Token Teachability in On-Policy Distillationcited by 42
  3. 2026Visual-Advantage On-Policy Distillation for Vision-Language Modelscited by 22
  4. 2026On-Policy Self-Distillation without Any Supervisioncited by 11

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026BAS-OPD: Budget-Aware Selective On-Policy Self-Distillation for Fine-Grained Multimodal Perception
  2. 2026Seeing Before Reasoning: Decoupling Perception and Reasoning for Shortcut-Resilient Multimodal On-Policy Self-Distillation
  3. 2026LEGO-OPD: Factorized Teacher Composition for Multimodal On-Policy Distillation
  4. 2026Visual-OPSD: Cross-Modal On-Policy Self-Distillation for Efficient Unified Multimodal Reasoning
  5. 2026Where-OPD: Spatially Guided On-Policy Self-Distillation of MLLMs with Synthetic Scenes
  6. 2026Distilling Visual Reasoning into Text Space
  7. 2026OPD-Aha: From Linguistic Momentum to Visual Reflection in Multimodal On-Policy Distillation
  8. 2026ViCuR: Visual Cues as Recoverable Privilege for Multimodal On-Policy Distillation