Unlocking the Potential of Anomaly Detection with the VAND 3.0 Challenge at CVPR
Join the VAND 3.0 CVPR Challenge! Tackle real-world anomalies, collaborate with experts, win prizes, and present at CVPR 2025.
Would you be ready to test your anomaly detection skills and make an impact? The CVPR VAND 3.0 Challenge is live! Co-organized by Voxel51, Anomalib, and MVTEC, this challenge invites developers, researchers, and innovators to push the boundaries of visual anomaly detection. With two exciting tracks — real-world visual inspection and few-shot learning in vision-language models — this is your chance to build something meaningful, win recognition, and shape the future of AI. Join us, bring your models, and let your ideas shine!
- Official challenge webpage
- Register now!
- Join our community (Channel: cvpr-challenge-vand3-0)
Introduction
The VAND 3.0 Challenge at CVPR is pushing the boundaries of anomaly detection, a critical technology with vast applications in industrial inspection and beyond. This year, Voxel51, Anomalib, and MVTec are co-organizing this exciting challenge, bringing the latest advancements in anomaly detection closer to real-world applications.
About the challenge
The VAND 3.0 Challenge consists of two categories aimed at advancing the field of visual anomaly detection:
Category 1: Focused on adapting and detecting robust anomalies in real-world visual inspection scenarios using the MVTEC Anomaly Detection Dataset. Participants will develop models based on a one-class training paradigm, using only normal images to demonstrate robustness against external factors and real-world variability.
Category 2: The VLN Anomaly Challenge, which focuses on few-shot learning and vision-language models to detect and localize structural and logical anomalies. This category encourages participants to use pre-trained models on public datasets, focusing on few-shot learning capabilities.
Why should I participate?
This challenge offers a unique opportunity to apply your skills to real-world problems. By participating, you’ll:
- Tackle real visual inspection challenges using industry-grade datasets like MVTec AD 2.
- Explore few-shot learning and vision-language models in the context of anomaly detection.
- Gain visibility in the CVPR community and receive feedback from top experts in the field.
- Compete for prizes and earn a spot to present your solution at our CVPR workshop.
- Join a vibrant community of developers and innovators pushing the boundaries of computer vision.
Workshop and community engagement
Over the past two years, our workshop has seen over 215 attendees participating both in person and online, with 100+ challenge teams actively engaging in the competition. For the 2024 edition, we received 40 submissions, with an acceptance rate of 62%, reflecting the high standards and competitive nature of the challenge. 2025 promises to be a great year as well.
Conclusion
The VAND 3.0 Challenge is not just about competition; it’s your chance to grow, connect, and make your mark in the anomaly detection landscape. VAND 3.0 is pushing the boundaries of what's possible in anomaly detection and fostering a community of innovators. We invite all developers to join us in this journey. Stay tuned to the Voxel51 blog and meetups for more insights and updates.
Want to participate?
Ready to take on the challenge and showcase your anomaly detection skills? Here’s how you can get involved:
Join our community: Connect with other developers and participants on our Discord.
Explore MVTEC AD2 with FiftyOne.
Submit your benchmark for Category 1: Visit the MVTEC Submission Platform for guidelines and submissions.
Submit your benchmark for Category 2: Head to the Anomalib Submission Platform to get started.
Connect on LinkedIn: Follow my journey and stay updated on LinkedIn.