PSBattles24K

该数据集包含了用户生成的篡改图像，其中包括7,000张原始图像和24,000张篡改后的变体。该数据集旨在评估图像哈希算法的性能。此外，数据集被划分为独立的训练集和测试集，其中训练集包含21,000对图像，测试集包含3,000对图像。另外，还创建了150,000张图像，用于评估对良性变换的不变性。规模上，数据集拥有7,000张原始图像和24,000张篡改后的变体，其任务是对图像哈希和归属进行评估。

This dataset contains user-generated tampered images, including 7,000 original images and 24,000 tampered variants. It is designed to evaluate the performance of image hashing algorithms. Moreover, the dataset is divided into independent training and test sets, where the training set contains 21,000 image pairs and the test set contains 3,000 image pairs. Additionally, 150,000 additional images were created to assess the invariance to benign transformations. In terms of scale, the dataset has 7,000 original images and 24,000 tampered variants, and its task is to evaluate both image hashing and image attribution.