MUIRBENCH
收藏资源简介:
MUIRBENCH是由南加州大学等机构创建的综合性多图像理解基准数据集,包含11,264张图片和2,600个多选题,覆盖12种多图像理解任务和10种多图像关系类别。数据集通过精心设计,每项任务都与一个不可回答的变体配对,以确保对多模态大语言模型进行可靠评估。数据来源多样,包括现有数据集、衍生数据和新收集数据,旨在全面评估模型在多图像场景下的理解和推理能力。该数据集的应用领域广泛,旨在推动模型超越单图像限制,提升对视觉世界的全面和集成理解。
MUIRBENCH is a comprehensive multi-image understanding benchmark dataset developed by institutions including the University of Southern California (USC). It comprises 11,264 images and 2,600 multiple-choice questions, covering 12 multi-image understanding tasks and 10 multi-image relationship categories. The dataset is meticulously engineered, with each task paired with an unanswerable variant to ensure reliable evaluation of multimodal large language models (LLMs). Drawing from diverse data sources including existing datasets, derived data and newly collected data, it aims to comprehensively evaluate a model's understanding and reasoning capabilities in multi-image scenarios. With a wide range of application domains, this dataset is designed to promote models to transcend the limitations of single-image understanding and enhance their holistic and integrated comprehension of the visual world.

- 1MuirBench: A Comprehensive Benchmark for Robust Multi-image Understanding南加州大学 · 2024年



