Env-Aware Anomaly Detection: Ignore Style Changes, Stay True to Content!

Abstract

We introduce a formalization and benchmark for the unsupervised anomaly detection task in the distribution-shift scenario. Our work builds upon the iWildCam dataset, and, to the best of our knowledge, we are the first to propose such an approach for visual data. We empirically validate that environment-aware methods perform better in such cases when compared with the basic Empirical Risk Minimization (ERM). We next propose an extension for generating positive samples for contrastive methods that considers the environment labels when training, improving the ERM baseline score by 8.7%.

PDF NeurIPSW OpenReview Semantic Scholar

Cite

Text

Smeu et al. "Env-Aware Anomaly Detection: Ignore Style Changes, Stay True to Content!." NeurIPS 2022 Workshops: DistShift, 2022.

Markdown

[Smeu et al. "Env-Aware Anomaly Detection: Ignore Style Changes, Stay True to Content!." NeurIPS 2022 Workshops: DistShift, 2022.](https://mlanthology.org/neuripsw/2022/smeu2022neuripsw-envaware/)

BibTeX

@inproceedings{smeu2022neuripsw-envaware,
  title     = {{Env-Aware Anomaly Detection: Ignore Style Changes, Stay True to Content!}},
  author    = {Smeu, Stefan and Burceanu, Elena and Nicolicioiu, Andrei Liviu and Haller, Emanuela},
  booktitle = {NeurIPS 2022 Workshops: DistShift},
  year      = {2022},
  url       = {https://mlanthology.org/neuripsw/2022/smeu2022neuripsw-envaware/}
}