Vision-Language Model Dialog Games for Self-Improvement
On the map: Multimodal LLMs / vision-language models → Embodied and 3D vision-language models
also in 1 more
Abstract (arXiv)
The increasing demand for high-quality, diverse training data poses a significant bottleneck in advancing vision-language models (VLMs). This paper presents VLM Dialog Games, a novel and scalable self-improvement framework for VLMs. Our approach leverages self-play between two agents engaged in a goal-oriented play centered around image identification. By filtering for successful game interactions, we automatically curate a high-quality dataset of interleaved images and text. We demonstrate that fine-tuning on this synthetic data leads to performance gains on downstream tasks and generalises across datasets. Moreover, as the improvements in the model lead to better game play, this procedure can be applied iteratively. This work paves the way for self-improving VLMs, with potential applications in various real-world scenarios especially when the high-quality multimodal data is scarce.
Key facts. Cited by 4 papers in the map (data through 7 Oct 2026). Placed in Embodied and 3D vision-language models, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 8 of its 26 references that are in the map. First posted on arXiv 4 Feb 2025.
Builds on
8 of 26 references.
Closest ancestors · 1
Same direction, newest first
- 2024Using Game Play to Investigate Multimodal and Conversational Grounding in Large Multimodal Models20 Jun
Other references · 7
- 2023GPT-4 Technical Reportcited by 18,132
- 2020Scaling Laws for Neural Language Modelscited by 6,411
- 2024Gemini 1.5: Unlocking multimodal understanding across millions of tokens of contextcited by 3,175
- 2022Constitutional AI: Harmlessness from AI Feedbackcited by 2,950
- 2025Multiagent Finetuning: Self Improvement with Diverse Reasoning Chainscited by 90
- 2024Synth2: Boosting Visual-Language Models with Synthetic Captions and Image Embeddingscited by 18
- 2023Leveraging Unpaired Data for Vision-Language Generative Models via Cycle Consistencycited by 8
Built on by
Citing papers per year: 2025: 3; 2026: 1.
4 papers.
- 2025Vision-Zero: Scalable VLM Self-Improvement via Strategic Gamified Self-Playcited by 37
- 2025Success and Cost Elicit Convention Formation for Efficient Communicationcited by 7
- 2025Self-Improvement in Multimodal Large Language Models: A Surveycited by 6
- 2026Dynamic Adversarial Reinforcement Learning for Robust Multimodal Large Language Modelscited by 0
- 2026Dynamic Adversarial Reinforcement Learning for Robust Multimodal Large Language Models24 Jan 2026
- 2025Success and Cost Elicit Convention Formation for Efficient Communication28 Oct 2025
- 2025Self-Improvement in Multimodal Large Language Models: A Survey3 Oct 2025
- 2025Vision-Zero: Scalable VLM Self-Improvement via Strategic Gamified Self-Play29 Sep 2025
Similar, not cited
Close in wording, no citation either way
- 2025VisPlay: Self-Evolving Vision-Language Models from Images
- 2025Game-RL: Synthesizing Multimodal Verifiable Game Data to Boost VLMs' General Reasoning
- 2025Are Large Vision Language Models Good Game Players?
- 2026MM-Zero: Self-Evolving Multi-Model Vision Language Models From Zero Data
- 2024ING-VP: MLLMs cannot Play Easy Vision-based Games Yet
- 2025V-MAGE: A Game Evaluation Framework for Assessing Vision-Centric Capabilities in Multimodal Large Language Models
- 2025Enhancing Vision-Language Model Training with Reinforcement Learning in Synthetic Worlds for Real-World Success
- 2024What matters when building vision-language models?