ArtiFact: Real and Fake Image Dataset

To maintain diversity across different sources, the dataset randomly samples real images from source datasets with numerous categories. Meanwhile, it generates synthetic images to match these categories. Furthermore, it uses captions and image masks from the COCO dataset to generate images via text-to-image and inpainting generators. Conversely, noise-to-image generators use normally distributed noise with various random seeds. To enhance realism, the dataset further processes the images to reflect real-world scenarios by applying random cropping, downscaling, and JPEG compression, all in accordance with IEEE VIP Cup 2022 standards.

The primary goal of the ArtiFact dataset is to serve as a benchmark for evaluating the performance of synthetic image detectors under real-world conditions. With its broad spectrum of diversity in terms of the generators used and levels of syntheticity, it provides a challenging environment for image detection tasks. Consequently, it pushes the boundaries of current detection capabilities.