# Visual Jigsaw Post-Training Improves MLLMs

arXiv: https://arxiv.org/abs/2509.25190 · HTML: https://mlsanity.com/paper/2509.25190/

## Key facts

- Cited by 26 papers in the map (data through 14 Oct 2026).
- Placed in Multimodal reasoning, one of 17 directions in LLM reasoning.
- First posted on arXiv 29 Sep 2025.

## Placed in

- LLM reasoning / [Multimodal reasoning](https://mlsanity.com/lr/multimodal-reasoning/)
- Multimodal LLMs / vision-language models / [Embodied and 3D vision-language models](https://mlsanity.com/mllm/embodied-and-3d-vision-language-models/)

## Built on by (most cited first, 26 of 26)

- [Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perception](https://mlsanity.com/paper/2602.11858/) (2026)
- [Spatial-SSRL: Enhancing Spatial Understanding via Self-Supervised Reinforcement Learning](https://mlsanity.com/paper/2510.27606/) (2025)
- [DiffThinker: Towards Generative Multimodal Reasoning with Diffusion Models](https://mlsanity.com/paper/2512.24165/) (2025)
- [Vero: An Open RL Recipe for General Visual Reasoning](https://mlsanity.com/paper/2604.04917/) (2026)
- [Revisiting the Necessity of Lengthy Chain-of-Thought in Vision-centric Reasoning Generalization](https://mlsanity.com/paper/2511.22586/) (2025)
- [ViSS-R1: Self-Supervised Reinforcement Video Reasoning](https://mlsanity.com/paper/2511.13054/) (2025)
- [CrystaL: Spontaneous Emergence of Visual Latents in MLLMs](https://mlsanity.com/paper/2602.20980/) (2026)
- [ReasonX: MLLM-Guided Intrinsic Image Decomposition](https://mlsanity.com/paper/2512.04222/) (2025)
- [PuzzleCraft: Exploration-Aware Curriculum Learning for Puzzle-Based RLVR in VLMs](https://mlsanity.com/paper/2512.14944/) (2025)
- [DiG: Differential Grounding for Enhancing Fine-Grained Perception in Multimodal Large Language Model](https://mlsanity.com/paper/2512.12633/) (2025)
- [When Thinking Hurts: Mitigating Visual Forgetting in Video Reasoning via Frame Repetition](https://mlsanity.com/paper/2603.16256/) (2026)
- [Unleashing the Intrinsic Visual Representation Capability of Multimodal Large Language Models](https://mlsanity.com/paper/2512.06281/) (2025)
- [Med-Scout: Curing MLLMs' Geometric Blindness in Medical Perception via Geometry-Aware RL Post-Training](https://mlsanity.com/paper/2601.23220/) (2026)
- [Learning Transferable Temporal Primitives for Video Reasoning via Synthetic Videos](https://mlsanity.com/paper/2603.17693/) (2026)
- [SSL-R1: Self-Supervised Visual Reinforcement Post-Training for Multimodal Large Language Models](https://mlsanity.com/paper/2604.20705/) (2026)
- [Dynamic Token Compression for Efficient Video Understanding through Reinforcement Learning](https://mlsanity.com/paper/2603.26365/) (2026)
- [OmniJigsaw: Enhancing Omni-Modal Reasoning via Modality-Orchestrated Reordering](https://mlsanity.com/paper/2604.08209/) (2026)
- [Boosting Visual Instruction Tuning with Self-Supervised Guidance](https://mlsanity.com/paper/2604.12966/) (2026)
- [JigShape: Evaluating Visual-Geometric Reasoning in VLMs through Jigsaw Puzzles](https://mlsanity.com/paper/2607.27670/) (2026)
- [The Art of Interrogation: Consistency Amplifies Factuality in Spatial Reasoning](https://mlsanity.com/paper/2606.11918/) (2026)
- [S2COPE: Self-Supervised Concept Discovery via Preference Learning](https://mlsanity.com/paper/2606.14586/) (2026)
- [LOCUS: Local Visual Cue Search for Enhancing Fine-Grained Perception in Multimodal Large Language Models](https://mlsanity.com/paper/2606.16586/) (2026)
- [The Shape of Time: Video-Token Contrast for Temporal Understanding in VideoLMs](https://mlsanity.com/paper/2609.04110/) (2026)
- [Same Reward, Different Skills: When Multimodal RL Learns to Look](https://mlsanity.com/paper/2610.01908/) (2026)
- [EventLens: Event-Structure Reinforcement Learning for Video Understanding](https://mlsanity.com/paper/2610.16172/) (2026)
- [Game2Real-Bench: Do Toy Games Reveal and Transfer to Real-World Visual Reasoning?](https://mlsanity.com/paper/2610.18834/) (2026)

## Abstract (arXiv)

Reinforcement learning based post-training has recently emerged as a powerful paradigm for enhancing the alignment and reasoning capabilities of multimodal large language models (MLLMs). While vision-centric post-training is crucial for enhancing MLLMs' intrinsic understanding of visual signals, current post-training paradigms are predominantly text-centric, where dense visual inputs are only leveraged to extract sparse cues for text-based reasoning. There exist a few approaches in this direction, however, they often still rely on text as an intermediate mediator or introduce additional visual generative designs. In this work, we introduce Visual Jigsaw, a generic self-supervised post-training framework designed to strengthen visual understanding in MLLMs. Visual Jigsaw is formulated as a general ordering task: visual inputs are partitioned, shuffled, and the model must reconstruct the visual information by producing the correct permutation in natural language. This naturally aligns with reinforcement learning from verifiable rewards (RLVR), requires no additional visual generative components, and derives its supervisory signal automatically without any annotations. We instantiate Visual Jigsaw across three visual modalities, including images, videos, and 3D data. Extensive experiments demonstrate substantial improvements in fine-grained perception, temporal reasoning, and 3D spatial understanding. Our findings highlight the potential of self-supervised vision-centric tasks in post-training MLLMs and aim to inspire further research on vision-centric pretext designs. Project Page: https://penghao-wu.github.io/visual_jigsaw/