Track 1: Adapt & Detect
This track introduces the MVTec AD 2 dataset, designed to advance anomaly detection algorithms under real-world conditions. It includes 8 new, challenging scenarios and enables testing against distribution shifts caused by varying lighting conditions. The focus lies on robustness and adaptability, encouraging models that can detect anomalies without prior knowledge of defects.
Track 2: VLM Anomaly Challenge
Here, participants work with the MVTec LOCO AD dataset to detect structural and logical anomalies in industrial products. This track is ideal for exploring few-shot learning and Vision-Language Models, tackling both physical defects and semantic inconsistencies.
Why participate?
- Showcase your innovative solutions in real-world anomaly detection
- Compete for prestigious prizes and present your work at the VAND Workshop at CVPR 2025
- Collaborate with leading companies and researchers in the field
Important dates
- MVTec AD 2 dataset release: April 1, 2025
- Challenge period: April 7 – May 26, 2025
- Results announced: June 3, 2025
For more details, visit the VAND 2025 website.