The “Synthetic Image Attribution Challenge” is organized within the Workshop on Deep Learn-
ing and Mathematical Methods for Deepfake Detection at the 35th International Conference on
Artificial Neural Networks (ICANN) 2026.
The challenge addresses the rapidly evolving problem of Synthetic Image source Attribution (SIA). Its
primary objective is to foster the development of novel, robust, and reproducible methodologies
for fake image attribution under controlled experimental conditions. In particular, the com-
petition will be conducted on a newly released ad hoc attribution dataset comprising images
generated by state-of-the-art text-to-image models.
Participants are invited to compete (alone or in team) in a SIA challenge carried out on a dataset of face images. the images are all synthetic and come from 10 different sources. The dataset is perfectly balanced among sources (all contribute the same number of images). Post-processing (operators will not be disclosed) has been applied to some test images.