I2EBench
收藏资源简介:
I2EBench是由厦门大学创建的一个全面的基于指令的图像编辑评估基准数据集。该数据集包含超过2000张图像和4000+条原始及多样化的编辑指令,旨在从多个维度评估图像编辑模型的质量。数据集的创建过程中,使用了ChatGPT来多样化指令,并通过现有的图像编辑模型生成编辑后的图像。I2EBench主要应用于图像编辑领域,旨在解决图像编辑模型的评估问题,提供一个全面的评估框架,以促进该领域的进一步发展。
I2EBench is a comprehensive instruction-based image editing evaluation benchmark dataset developed by Xiamen University. This dataset includes over 2,000 images and more than 4,000 original and diverse editing instructions, which is designed to evaluate the quality of image editing models from multiple dimensions. During the dataset creation process, ChatGPT was utilized to diversify the editing instructions, and existing image editing models were employed to generate the edited images. Primarily applied in the field of image editing, I2EBench aims to address the evaluation challenges of image editing models by providing a comprehensive assessment framework, so as to facilitate further advancements in this domain.

- 1I2EBench: A Comprehensive Benchmark for Instruction-based Image Editing厦门大学 · 2024年



