Introduction

The 2025 Adversarial Attack Challenge for Secure Face Recognition (AAC) aims to enhance the robustness and security of facial recognition (FR) systems against adversarial attacks. With the widespread adoption of face recognition technology in critical applications such as security systems, financial authentication, and border control, adversarial attacks present a significant threat to the reliability, security, and trustworthiness of these systems. These attacks exploit the vulnerabilities of deep learning models by subtly altering input images, leading to severe consequences like unauthorized access and identity fraud.

The challenge purpose is to foster the collaboration among experts in machine learning, cybersecurity, and biometric authentication, contributing to the development of more resilient and secure face recognition systems. Therefore, the AAC introduces two main tracks to evaluate both adversarial classification and recognition impact: (1) The Detection Track, focused on developing models that accurately classify face images as either adversarial or clean, and (2) The Resilience Track, which challenges participants to create FR models that maintain high performance even when faced with adversarially manipulated images

The Challenge

The Adversarial Attack Challenge (AAC) explores the use of adversarial attacks to improve the robustness of Face Recognition (FR) systems.
The challenge has two tracks:

Detection Track

• Objective: Develop a model that accurately classifies face images as either “clean” or “adversarial.”.
• Dataset Composition: Develop a model that accurately classifies images as either "adversarial" or "clean."

Resilience Track

• Objective: Train a face recognition model that remains robust under adversarial attacks.
• Dataset Composition: Contains both clean images and adversarially altered face images, including various types of attack modifications.

Challenge Rules

Teams can choose to participate in one or both tracks. In both tracks solution will be required to be open-source and their key focus should be the generalization capability of the models. After evaluation, performance metrics will be released, and top-performing teams will be recognized. Top 3 teams will be invited to co-author a paper summarizing the challenge results and will be asked to provide brief explanations of their solutions.

Awards

For each track a monetary price will be attributed to the top 3 performing teams.

The submissions must comply with challenge rules and outperform the baseline methods.

The prizes are:

• 1st place - $1000
• 2nd place - $500
• 3rd place - $250

Important Dates

• March 17, 2025 - Github instruccions release.
• March 17, 2025 - Adversarial attack package release.
• March 18, 2025 - Dataset release.
• May 31, 2025 - Deadline for Algorithm evaluation on platform.
• June 10, 2025 - Announcement of the results to participants.
• June 23, 2025 - Submission of summary papers.
• July 23, 2025 - Camera-ready papers.
• September 8-11, 2025 - IJCB conference.

Submission

To submit your solutions, please send the submission content (as noted on the the github page: https://github.com/dev-yoonik/IJCB-AAC-2025/blob/main/AAC2025_Submission/README.md) as a zipped folder or cloud drive link, to the organizers email at adversarial@youverse.id

Register Now

Secure your spot in the challenge by registering today!

Organization

Organized by Youverse and the Institute of Systems and Robotics, University of Coimbra

This competition is sponsored by:

Contact Us

For inquiries, reach out to us at: adversarial@youverse.id