IF-Bench is a high-quality benchmark for multimodal understanding of infrared images, jointly released in 2025 by the Institute of Automation, Chinese Academy of Sciences, and the School of Artificial Intelligence, University of Chinese Academy of Sciences. Related papers include... IF-Bench: Benchmarking and Enhancing MLLMs for Infrared Images with Generative Visual PromptingThe aim is to systematically evaluate the semantic understanding capabilities of multimodal large language models (MLLMs) for infrared images.

This dataset contains 499 infrared images and 680 visual question-and-answer (VQA) pairs. The images are sourced from 23 different infrared image datasets, maintaining a relatively balanced distribution. The dataset is built around an infrared image understanding task, covering 10 key dimensions of image understanding. All questions are provided in both Chinese and English, with the order of the options randomly shuffled to ensure that the correct answers are evenly distributed across options A–D.