TheGarlic/image-authenticity-battle

Image Authenticity Battle Dataset包含真实和合成/篡改的图像，用于人工智能生成内容检测的人类感知研究。数据集共有19500张图像，分为真实、合成（完全由AI生成）和篡改（AI编辑）三个类别。数据集中的图像由Nano Banana、Qwen、Flux和SD3等模型生成或编辑。每个图像都附带元数据，包括文件路径、数据集名称、类别、使用模型、详细类型信息、详细排行榜的ELO评分和模型排行榜的ELO评分。该数据集旨在用于基于ELO评分系统的图像对比战斗系统。

The Image Authenticity Battle Dataset includes real and synthetic/tampered images for human perception studies on AI-generated content detection. The dataset consists of 19,500 images across three categories: Real, Synthetic (Fully AI-generated), and Tampered (AI-edited). The images in the dataset are generated or edited using models such as Nano Banana, Qwen, Flux, and SD3. Each image is accompanied by metadata including the file path, dataset name, category, model used, detailed type information, ELO rating for the detailed leaderboard, and ELO rating for the model leaderboard. This dataset is designed for use in ELO-based battle systems where users compare real versus synthetic/tampered images.