Same Same But DifferNet: Semi-Supervised Defect Detection with Normalizing Flows

28 Aug 2020 Marco Rudolph Bastian Wandt Bodo Rosenhahn

The detection of manufacturing errors is crucial in fabrication processes to ensure product quality and safety standards. Since many defects occur very rarely and their characteristics are mostly unknown a priori, their detection is still an open research question... (read more)

PDF Abstract

Datasets


Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
RESULT BENCHMARK
Anomaly Detection MVTec AD DifferNet Detection AUROC 94.9 # 7
Anomaly Detection Surface Defect Saliency of Magnetic Tile DifferNet (unsupervised) Detection AUROC 97.7 # 1

Methods used in the Paper


METHOD TYPE
Affine Coupling
Bijective Transformation
Batch Normalization
Normalization
RealNVP
Generative Models
Normalizing Flows
Distribution Approximation
Adam
Stochastic Optimization
DifferNet
Semi-Supervised Learning Methods