MULTI-Benchmark：带有文本和图像的多模态理解排行榜

该数据集为上海交通大学发布的多模态基准测试 MULTI ，旨在评测多模态大模型在理解复杂的表格和图像、以及进行长文本推理的能力。该测试提供多模态输入，并要求回答是精确的或开放式的，反映了现实生活中的考试风格。 MULTI 包含逾 18,000 个问题，涵盖从公式推导到图像分析及跨模态推理的多样化任务。

This dataset is the MULTI multimodal benchmark released by Shanghai Jiao Tong University, which is designed to evaluate the capabilities of multimodal large models in understanding complex tables and images, as well as performing long-text reasoning. This benchmark provides multimodal inputs, and requires responses to be either exact or open-ended, mirroring the styles of real-life examinations. The MULTI benchmark contains over 18,000 questions, covering diverse tasks ranging from formula derivation through image analysis to cross-modal reasoning.

搜集汇总

背景与挑战

背景概述

MULTI-Benchmark是由上海交通大学发布的多模态基准测试数据集，包含超过18,000个问题，涵盖公式推导、图像分析及跨模态推理等多样化任务，旨在评测多模态大模型在复杂表格和图像理解以及长文本推理方面的能力。数据集还包括MULTI-Elite和MULTI-Extend两个子集，分别针对难题和外部知识上下文进行评测。