Mitigating the Bias of Centered Objects in Common Datasets

TL;DR

This paper identifies a bias in common datasets where objects are over-represented at the center, affecting CNN performance near image boundaries, and proposes data augmentation to mitigate this issue.

Contribution

It reveals the central object bias in datasets and demonstrates how data augmentation can reduce its impact on CNN accuracy.

Findings

Object bias affects CNN performance at image boundaries

Data augmentation mitigates boundary-related accuracy drops

Bias is prevalent in commonly used datasets

Abstract

Convolutional networks are considered shift invariant, but it was demonstrated that their response may vary according to the exact location of the objects. In this paper we will demonstrate that most commonly investigated datasets have a bias, where objects are over-represented at the center of the image during training. This bias and the boundary condition of these networks can have a significant effect on the performance of these architectures and their accuracy drops significantly as an object approaches the boundary. We will also demonstrate how this effect can be mitigated with data augmentation techniques.