This dataset is the image stimulus pool of 50 deepfake and 50 real images, used for the experiment in the study titled "Testing Human Ability To Detect Deepfake Images of Human Faces".

Images were obtained through random selection from the Flickr Faces High Quality dataset (https://paperswithcode.com/dataset/ffhq) and likewise from output of the StyleGAN2 algorithm (https://paperswithcode.com/method/stylegan2) as trained on the FFHQ dataset.

Also included are the 20 deepfake images, similarly obtained (from StyleGAN2:FFHQ), which were used in the familiarization intervention in the study; and the 20 deepfake images (different images but similarly obtained, albeit with an element of curation to select for images with the specific "tell-tale signs" / "visible rendering artefacts") which were used in the advice intervention in the study.

The 50 real and 50 deepfake images that were used as test stimuli in the experiment are in /real and /fake respectively; the 20 familiarization images are in /familiarization; and the 20 images used in the advice intervention are in /advice.

To comprehensively evaluate our UniFaceForge model across all types of face forgery attacks, we created a diverse dataset combining real face images from the CelebA, FFHQ datasets and manipulated images including traditional Face morphing, Face swapping with varying blending degrees, regional retouching and face attributes shifting, GAN-based alterations such as StyleGAN2-based attribute modifications, StarGAN v2 style transfer, STGAN-generated identity transformations, and advanced synthetic… See the full description on the dataset page: https://huggingface.co/datasets/TheDeepDas/UniFaceForge-Real-Images.