第六届全球校园人工智能算法精英大赛AI 生成人脸图像鉴别赛题数据集B

主办方提供测试数据集，包含5000~6000张人脸图像。该数据集中包含由Stable Diffusion为主的不同AI工具网络生成的AI生成人脸图像，以及主办方自行收集的或从互联网中爬取得到的原始人脸图像。每张图像可能包含单人人脸或多人人脸，可能是自然人脸也可能是非自然人脸（如卡通、素描），但有任意一部分（人脸部分或非人脸部分）为AI生成即应将该图像视为AI生成图像。数据集中AI生成人脸图像和原始人脸图像各约占一半，所有图像文件均为jpg格式，由主办方以16位随机字符（字母+数字）命名，像素大小>=128*128，但清晰度各不相同。

The organizer provides a test dataset containing 5,000 to 6,000 facial images. This dataset includes two types of samples: AI-generated facial images created by various AI tools and networks with Stable Diffusion as the primary representative, and original facial images collected by the organizer or crawled from the public Internet. Each image may contain one or multiple faces, and can be either a natural human face or a non-natural face such as a cartoon or sketch. Any image with at least one AI-generated component, whether in the facial region or non-facial areas, should be categorized as an AI-generated image. The dataset has approximately equal proportions of AI-generated facial images and original facial images. All image files are in JPG format, named by the organizer using 16 random alphanumeric characters (letters and digits), with a resolution of no less than 128 × 128 pixels, but varying in image clarity.

搜集汇总

背景与挑战

背景概述

该数据集是第六届全球校园人工智能算法精英大赛中用于AI生成人脸图像鉴别的测试集，包含约5000至6000张人脸图像。其中AI生成图像与原始图像各占一半，图像均为jpg格式且像素不低于128*128，但清晰度存在差异。