NexViTAD: Few-shot Unsupervised Cross-Domain Defect Detection via Vision Foundation Models and Multi-Task Learning

2025-07-10

AI Summary

To address domain shift in industrial few-shot unsupervised cross-domain anomaly detection, this paper proposes NexViTAD—a novel framework integrating vision foundation models (Hiera/DINO-v2). It introduces a hierarchical adapter for parameter-efficient domain adaptation, a shared subspace projection to align feature distributions across domains, and a multi-task decoder for joint reconstruction and discrimination. Anomaly scoring is driven by Sinkhorn-K-means clustering, enhancing separation of normal and anomalous patterns. Feature robustness is improved via bottleneck-dimensional constraints, skip connections, and Gaussian filtering; pixel-level localization is achieved through adaptive thresholding. Evaluated on MVTec AD, NexViTAD achieves 97.5% AUC, 70.4% AP, and 95.2% PRO—substantially outperforming state-of-the-art methods. The framework establishes an efficient, transferable paradigm for cross-domain industrial defect detection.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

This paper presents a novel few-shot cross-domain anomaly detection framework, Nexus Vision Transformer for Anomaly Detection (NexViTAD), based on vision foundation models, which effectively addresses domain-shift challenges in industrial anomaly detection through innovative shared subspace projection mechanisms and multi-task learning (MTL) module. The main innovations include: (1) a hierarchical adapter module that adaptively fuses complementary features from Hiera and DINO-v2 pre-trained models, constructing more robust feature representations; (2) a shared subspace projection strategy that enables effective cross-domain knowledge transfer through bottleneck dimension constraints and skip connection mechanisms; (3) a MTL Decoder architecture supports simultaneous processing of multiple source domains, significantly enhancing model generalization capabilities; (4) an anomaly score inference method based on Sinkhorn-K-means clustering, combined with Gaussian filtering and adaptive threshold processing for precise pixel level. Valuated on the MVTec AD dataset, NexViTAD delivers state-of-the-art performance with an AUC of 97.5%, AP of 70.4%, and PRO of 95.2% in the target domains, surpassing other recent models, marking a transformative advance in cross-domain defect detection.

Problem

• Addresses domain-shift in industrial anomaly detection
• Enables cross-domain knowledge transfer via shared subspace
• Improves defect detection with multi-task learning

Innovation

• Hierarchical adapter fuses Hiera and DINO-v2 features
• Shared subspace projection enables cross-domain transfer
• MTL Decoder processes multiple source domains simultaneously