Flow Mismatching: Unsupervised Anomaly Detection via Velocity Discrepancies in Flow Matching Models

Shengzhe Chen, Mehrdad Moradi, Kamran Paynabar, Hao Yan

arXiv:2605.23070 · 2026-05-25 공개 · arXiv · PDF

flow-matching unsupervised-learning anomaly-detection denoising mvtec-ad fisher-divergence flow-mismatching velocity-discrepancies

Abstract

We propose Flow Mismatching, an unsupervised anomaly detection method that deliberately avoids reconstruction-based paradigms. Instead, we treat flow matching as geometric dynamics and leverage a key insight: anomalies occur at places where the learned normal flow disagrees with the geometric path toward a test image. Given a flow matching model trained only on normal images, we probe its learned velocity field along affine paths from Gaussian noise to a target image. Along each path, we compare the model-predicted velocity, which follows normal generative dynamics, with the geometric velocity toward the target, which includes any anomalous content. Anomalies induce strong local disagreement between these velocities. Aggregating the mismatch over different time steps and multiple paths yields pixel-wise heatmaps and image-level scores without test-time optimization, feature memories, or additional calibration. Our analysis shows that the population mismatch decomposes into an irreducible denoising term and a Fisher-divergence term between the test-path and normal-path score functions, which identifies the score-gap component that drives anomaly separation and explains the effectiveness of robust path aggregation. Extensive experiments on MVTec-AD and VisA demonstrate superior performance compared with SOTA reconstruction-based and recent flow matching-based approaches.

한국어 요약

📋 한 줄 요약

**[비지도 이상 탐지 / Flow Matching]** Flow Mismatching이 정상 학습된 flow matching의 학습 velocity와 target 방향 geometric velocity의 affine path mismatch로 anomaly 탐지, MVTec-AD·VisA에서 SOTA reconstruction·flow 기반 능가.

🎯 핵심 기여도

💡 핵심 아이디어

정상 이미지로 학습된 flow matching model의 velocity field가 anomalous content를 포함한 target으로의 geometric velocity와 disagree하는 지점에 anomaly가 위치하며, affine path를 따라 velocity mismatch를 측정·집계하면 test-time optimization 없이 anomaly가 자연스럽게 분리된다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: Reconstruction paradigm 의존 없는 새로운 flow matching 기반 anomaly detection, 이론·실험 양면 결합, test-time overhead 없는 inference로 실용성. **한계**: Affine path 가정의 표현력 의존, multiple path 집계의 계산 비용, 매우 미세·structural anomaly에 대한 sensitivity 검증 필요.

🚀 실용적 활용