Deep Nearest Neighbor Anomaly Detection
This work provides a practical solution for anomaly detection tasks, showing that simple methods can be more effective than complex self-supervised techniques, but it is incremental as it builds on existing nearest-neighbor and feature-based approaches.
The paper tackled the problem of anomaly detection by comparing nearest-neighbor methods using Imagenet pre-trained features against self-supervised deep methods, finding that the nearest-neighbor approach outperforms in accuracy, few-shot generalization, training time, and noise robustness.
Nearest neighbors is a successful and long-standing technique for anomaly detection. Significant progress has been recently achieved by self-supervised deep methods (e.g. RotNet). Self-supervised features however typically under-perform Imagenet pre-trained features. In this work, we investigate whether the recent progress can indeed outperform nearest-neighbor methods operating on an Imagenet pretrained feature space. The simple nearest-neighbor based-approach is experimentally shown to outperform self-supervised methods in: accuracy, few shot generalization, training time and noise robustness while making fewer assumptions on image distributions.