Anomaly Detection
Nearest neighbor (kNN) methods utilizing deep pre-trained features exhibit very strong anomaly detection performance when applied to entire images. A limitation of kNN methods is the lack of segmentation map describing where the anomaly lies inside the image.
SPADE(Semantic Pyramid Anomaly Detection) tackles this challenge.
Architecture
- Feature Extraction
The paper uses the Resnet Feature Extracter pre-trained on ImageNet. - KNN Normal Image Retrieval
Determine which images contain anomalies by measuring the distance between image-level features and checking whether it exceeds a certain threshold. - Sub-Image Anomaly Detection
Construct a “Gallery” of features at all pixel locations of the K nearest neighbour. The anomaly score at pixel p, is given by the average distance between the features F(y, p) and its κ nearest features from the gallery G
