Vision-G1: Towards General Vision Language Reasoning with Multi-Domain Data Curation

TL;DR

Vision-G1 introduces a comprehensive multi-domain visual reasoning dataset and employs a multi-round RL training approach, significantly enhancing general reasoning abilities across diverse tasks and outperforming existing models.

Contribution

The paper presents a large, multi-domain dataset and a novel influence function-based data selection method for training a general visual language model with reinforcement learning.

Findings

Achieves state-of-the-art results on various visual reasoning benchmarks.

Outperforms similar-sized VLMs and proprietary models like GPT-4o and Gemini-1.5 Flash.

Demonstrates effective multi-domain generalization in visual reasoning.

Abstract

Despite their success, current training pipelines for reasoning VLMs focus on a limited range of tasks, such as mathematical and logical reasoning. As a result, these models face difficulties in generalizing their reasoning capabilities to a wide range of domains, primarily due to the scarcity of readily available and verifiable reward data beyond these narrowly defined areas. Moreover, integrating data from multiple domains is challenging, as the compatibility between domain-specific datasets remains uncertain. To address these limitations, we build a comprehensive RL-ready visual reasoning dataset from 46 data sources across 8 dimensions, covering a wide range of tasks such as infographic, mathematical, spatial, cross-image, graphic user interface, medical, common sense and general science. We propose an influence function based data selection and difficulty based filtering strategy to identify high-quality training samples from this dataset. Subsequently, we train the VLM, referred to as Vision-G1, using multi-round RL with a data curriculum to iteratively improve its visual reasoning capabilities. Our model achieves state-of-the-art performance across various visual reasoning benchmarks, outperforming similar-sized VLMs and even proprietary models like GPT-4o and Gemini-1.5 Flash. The model, code and dataset are publicly available at https://github.com/yuh-zha/Vision-G1.