VisualSphinx: Large-Scale Synthetic Vision Logic Puzzles for RL
This addresses a data bottleneck for researchers and developers working on VLM reasoning, though it is incremental as it focuses on synthetic data generation rather than a new model paradigm.
The paper tackles the lack of large-scale, structured training data for vision language models (VLMs) in multimodal reasoning by proposing VisualSphinx, a synthetic dataset of visual logic puzzles, and shows that VLMs trained on it exhibit improved performance on logical reasoning tasks.
Vision language models (VLMs) are expected to perform effective multimodal reasoning and make logically coherent decisions, which is critical to tasks such as diagram understanding and spatial problem solving. However, current VLM reasoning lacks large-scale and well-structured training datasets. To bridge this gap, we propose VisualSphinx, a first-of-its-kind large-scale synthetic visual logical reasoning training data. To tackle the challenge of image synthesis with grounding answers, we propose a rule-to-image synthesis pipeline, which extracts and expands puzzle rules from seed questions and generates the code of grounding synthesis image synthesis for puzzle sample assembly. Experiments demonstrate that VLM trained using GRPO on VisualSphinx benefit from logical coherence and readability of our dataset and exhibit improved performance on logical reasoning tasks. The enhanced reasoning capabilities developed from VisualSphinx also benefit other reasoning tasks such as algebraic reasoning, arithmetic reasoning and geometry reasoning.