R1-SyntheticVL: Is Synthetic Data from Generative Models Ready for Multimodal Large Language Model?

In this work, we aim to develop effective data synthesis techniques that autonomously synthesize multimodal training data for enhancing MLLMs in solving complex real-world tasks. To this end, we propose Collective Adversarial Data Synthesis (CADS), a novel and general approach to synthesize high-quality, diverse and challenging multimodal data for MLLMs. The core idea of CADS is to leverage collective intelligence to ensure high-quality and diverse generation, while exploring adversarial learning to synthesize challenging samples for effectively driving model improvement. Specifically, CADS operates with two cyclic phases, i.e., Collective Adversarial Data Generation (CAD-Generate) and Collective Adversarial Data Judgment (CAD-Judge). CAD-Generate leverages collective knowledge to jointly generate new and diverse multimodal data, while CAD-Judge collaboratively assesses the quality of synthesized data. In addition, CADS introduces an Adversarial Context Optimization mechanism to optimize the generation context to encourage challenging and high-value data generation. With CADS, we construct MMSynthetic-20K and train our model R1-SyntheticVL, which demonstrates superior performance on various benchmarks.

Modern AI systems often need large amounts of carefully prepared training data, but such data are expensive to collect and label. This paper proposes Collective Adversarial Data Synthesis (CADS), a way to automatically create new image-text training examples. CADS uses a group of AI models to generate candidate examples and then check whether they are useful, varied, and reliable. It also encourages the system to create harder examples, so that the trained model can learn from cases that better test its reasoning ability. Using this method, the authors build a dataset called MMSynthetic-20K and train a model called R1-SyntheticVL. The model performs strongly on several standard tests, showing that well-designed synthetic data can help improve multimodal AI systems while reducing the need for costly human-labeled data.