docci

DOCCI是一个图像数据集，包含约1万到10万张配有详细描述的图像，这些描述解释了图像的关键元素以及背景、光照和设置等辅助信息。该数据集专门用于评估图像的精确视觉属性，并包含许多具有关键差异的相关图像。所有描述都经过人工标注，以确保它们能够充分区分每张图像与其对应的图像。DOCCI支持文本生成图像和图像生成文本的任务，并采用CC-BY 4.0授权许可。

DOCCI is an image dataset containing approximately 10,000 to 100,000 images, each paired with detailed descriptions that explain the key elements of the image as well as supplementary information such as background, lighting, and scene settings. This dataset is specifically designed for evaluating the precise visual attributes of images, and includes a large number of related images with critical differences between them. All descriptions are manually annotated to ensure that they can sufficiently distinguish each individual image from its corresponding counterparts. DOCCI supports both text-to-image and image-to-text generation tasks, and is licensed under CC-BY 4.0.

搜集汇总

背景与挑战

背景概述

DOCCI是一个包含约1万到10万张图像的数据集，每张图像都配有详细的人工标注描述，这些描述解释图像的关键元素和辅助信息，旨在精确评估视觉属性并区分相关图像。该数据集支持文本到图像和图像到文本的生成任务，采用CC-BY 4.0许可，语言为英语。