Competition Description
Competition Description
This version of the IJCB PAD ID-Card 2026 will have two tracks:
Track 1—Algorithms: Participant teams in this track may use only the provided dataset for training. This PAD system considers ID cards and Passports from random template-generated documents printed on PVC as bona fide, and attacks such as print, screen, and composite are based on this PVC template. This approach is aligned with the open sets available in the state of the art for research purposes.
A new training set for 2026 will be available only for registered teams. This track will explore and test various algorithms, seeking the best solutions for ID cards and passports.
Track 2—Research: This is a PAD system where the bona fide is a genuine ID card from volunteer participants. Under this condition, a PVC is considered a print attack. All the attacks (print, screen, and composite) are generated from the genuine ID card. This approach is aligned with the commercial and proprietary datasets and the results of the previous competition.
This track is fully open to academic partners and companies because it represents the most realistic test in the SOTA.
Track 2 can be complemented with state-of-the-art sets, such as KID34K, IDNET, and others, which have more than 100GB of ID card and passport images.
The test set will be sequestered and unavailable to all the participants. This dataset contains bona fide, composite* (manual and automatic methods), print, and screen versions for at least 4 ID card countries (ICAO- and non-ICAO-compliant).
No injection attack will be tested.
(*) The composite attacks involve swapping faces and text areas on ID cards.
In the competition, the test set will not be pre-processed it is up to each team to implement the pre-process in their docker image.
The test will be performed in stand-alone conditions, which means without an internet connection.
*Bona fide: For track 1, the *the simulated-bona fide images were created from empty templates filled with random data.
*Bona fide: For track 2, the *bona fide image is a genuine capture without any modification.
Printed: Represent the printed version of *bona fide with different printing machines
Screen: Represent the *bona fide image displayed on different monitors, tablets, and TVs.
Composite: Represent one *bona fide image with copy and paste areas from the face and the text.
As in the previous competition (2025), in track 2, the participants can use any proprietary, homemade, synthetic, or open-access dataset available in the state of the art for the training process, such as MIDV500, DLC2021, KID34K, IDNet, and others.
Examples of some open-access datasets are described here:
D. Benalcazar, J. E. Tapia, S. Gonzalez and C. Busch, "Synthetic ID Card Image Generation for Improving Presentation Attack Detection," in IEEE Transactions on Information Forensics and Security, vol. 18, pp. 1814-1824, 2023, doi: 10.1109/TIFS.2023.3255585.
Arlazarov, V.V.; Bulatov, K.; Chernov, T.; Arlazarov, V.L. MIDV-500: A Dataset for Identity Document Analysis and Recognition on Mobile Devices in Video Stream. Comput. Opt. 2019, 43, 818–824.
Polevoy DV, Sigareva IV, Ershova DM, Arlazarov VV, Nikolaev DP, Ming Z, Luqman MM, Burie J-C. Document Liveness Challenge Dataset (DLC-2021). Journal of Imaging. 2022; 8(7):181. https://doi.org/10.3390/jimaging8070181.
R. P. Markham, J. M. E. López, M. Nieto-Hidalgo and Juan. E. Tapia, "Open-Set: ID Card Presentation Attack Detection Using Neural Style Transfer," in IEEE Access, vol. 12, pp. 68573-68585, 2024, doi: 10.1109/ACCESS.2024.3397190
PARK, E.-J., WOO, S. S., & Yoon, K. (2023). KID34K: A Dataset for Identity Card Fraud Detection [Data set]. Zenodo. https://doi.org/10.5281/zenodo.8034016
IDNet: Identity Analysis Image Dataset, A Novel Passport Dataset for Identity Document Analysis and Fraud Detection. https://www.kaggle.com/datasets/chitreshkr/idnet-identity-document-analysis.
**The test set sequestered will be common for both tracks 1 and 2.**