Self-supervised augmenting AnomalyCLIP

Abstract

In order to better solve the problem of poor robustness of the current anomaly detection model in the real scene.We use data augmentation and anomaly synthesis techniques to assist in training AnomalyCLIP which performs well on zero-shot detection.Our method compared with the baseline, the performance of the original MVTec dataset is improved.

Introduction

Other relevant content is coming soon

Background:

Challenge Description:

Methodology

Model Design

Dataset & Evaluation

Results

Discussion

Conclusion

After auxiliary training with our proposed self-supervised method, performance improvement has been achieved in the original MVTec, and we believe that our method can show stronger robustness in more varied real-world scenarios