Abnormal Object Recognition: A Comprehensive Study

TL;DR

This paper presents a new approach to abnormal object recognition by modeling typicalities and deviations, introducing a dataset, and demonstrating how to identify and explain abnormalities in images.

Contribution

It introduces a novel abnormality detection model, a dedicated dataset, and methods for recognizing and explaining abnormalities in visual recognition tasks.

Findings

Effective abnormality detection demonstrated on the new dataset

Model can identify main reasons for abnormalities

Insights into reasoning about deviations from typical object properties

Abstract

When describing images, humans tend not to talk about the obvious, but rather mention what they find interesting. We argue that abnormalities and deviations from typicalities are among the most important components that form what is worth mentioning. In this paper we introduce the abnormality detection as a recognition problem and show how to model typicalities and, consequently, meaningful deviations from prototypical properties of categories. Our model can recognize abnormalities and report the main reasons of any recognized abnormality. We introduce the abnormality detection dataset and show interesting results on how to reason about abnormalities.