Reproduce “SimpleNet: A Simple Network for Image Anomaly Detection and Localization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhikang Liu, Yiming Zhou, Yuansheng Xu, Zilei Wang
SimpleNet introduces a streamlined neural network architecture for unsupervised image anomaly detection and localization, achieving state-of-the-art performance on the MVTec AD benchmark with 99.6% image-level AUROC and an inference speed of 77 frames per second. This approach uniquely combines high accuracy with exceptional efficiency, addressing a critical need for real-time industrial quality control applications.
Key insights
- Mitigating domain bias with a simple Feature Adaptor (single FC layer) can significantly improve the compactness and discriminability of learned feature representations for anomaly detection.
- Synthesizing anomalous features by adding controlled Gaussian noise in the latent space is more effective than generating anomalies in pixel space, leading to a tighter normal feature distribution boundary.
- High accuracy in image anomaly detection can be achieved with a simple, efficient network design and a shallow discriminator, enabling remarkable inference speeds for real-time applications.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.