DCkNN is a general pipeline for anomaly detection in images. It used multiple activation maps that are given from a pre-trained ImageNet based neural network like ResNet, finds the k nearest neighbours of some test sample from every set of activations and uses a majority vote on all those to classify the sample as "normal" or "anomalous"
