Measuring Aesthetic Information in AI-Generated Artistic Images: The Dataset

This dataset contains the human and AI-generated artworks used in the study presented in the paper \Measuring Aesthetic Information in AI-Generated Artistic Images: A Tool for Authorship Detection\. It is a curated collection comprising 994 human artworks and their corresponding 994 AI-generated images, produced using Stable Diffusion v2.1. The number of paintings was determined by selecting, from the WikiArt database, human artworks whose dimensions closely matched the native size of the images generated by Stable Diffusion v2.1 (768\u00d7768 pixels). Specifically, all paintings dated from 1800 onward were considered, and a tolerance range of \u00b120% was applied, resulting in the inclusion of artworks with dimensions approximately between 614\u00d7614 and 922\u00d7922 pixels. In the aforementioned paper, any residual difference in image size was addressed by resizing the AI-generated images to match the dimensions of their human counterparts. All images are compressed in JPG format. Since the images sourced from WikiArt are already in JPG, the AI-generated images were also compressed to JPG using the same quality factor as their corresponding human counterparts.