We’re hosting CircleID, the ICDAR 2026 competition on learning subtle details in hand-drawn circles from scanned images. The challenge features two tracks: Writer Identification and Pen Classification.
CircleID includes an out-of-distribution twist for Writer Identification: The test set contains circles from writers not seen during training, so methods must learn to recognize unknown writers.
To help participants get started quickly, we provide a minimal ResNet-baseline that runs out of the box and is easy to tune.
Baseline code: GitHub
Kaggle competitions: Writer Identification · Pen Classification
Challenge homepage: Project page
Organizers: Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Lehrstuhl für Mustererkennung (Informatik 5), and Johannes Gutenberg-Universität Mainz (JGU), Buchwissenschaft.