ReasonMap Traffic Graph Reasoning Benchmark Dataset

This dataset is a new evaluation benchmark proposed by a team from Westlake University, National University of Singapore, Zhejiang University, and Huazhong University of Science and Technology in 2025. The relevant paper results are:Can MLLMs Guide Me Home? A Benchmark Study on Fine-Grained Visual Reasoning from Transit Maps", ReasonMap emphasizes spatial relationships and route reasoning in images. It is the first multimodal reasoning evaluation benchmark focusing on high-resolution traffic maps (mainly subway maps) and is designed to evaluate the ability of large models to understand fine-grained structured spatial information in images.

Dataset features:

• High-resolution challenge: The average resolution of each map image in the dataset is as high as 5839 × 5449, which is much higher than existing visual reasoning tasks, and places higher requirements on the image encoding capabilities of the model.

• Difficulty-aware design: Images are labeled with difficulty to ensure a balanced distribution of question-answer pairs at different difficulty levels, helping to more comprehensively evaluate model capabilities.

• Multi-dimensional evaluation system: not only examines the accuracy of the model's answers, but also conducts a fine-grained evaluation of the quality of the model route, including path rationality and transfer strategies.

• Close to real-world usage scenarios: The tasks are directly based on image reasoning, do not rely on structured middleware, and are closer to the way humans think when using maps.