Texture Bias Of CNNs Limits Few-Shot Classification Performance
This addresses the challenge of accurate image classification with limited labeled data for computer vision researchers, offering an incremental improvement by correcting a known bias.
The study tackled the problem of few-shot image classification by investigating how the texture bias in CNNs harms performance, and after correcting this bias, they achieved state-of-the-art results on the miniImageNet task with a simpler method.
Accurate image classification given small amounts of labelled data (few-shot classification) remains an open problem in computer vision. In this work we examine how the known texture bias of Convolutional Neural Networks (CNNs) affects few-shot classification performance. Although texture bias can help in standard image classification, in this work we show it significantly harms few-shot classification performance. After correcting this bias we demonstrate state-of-the-art performance on the competitive miniImageNet task using a method far simpler than the current best performing few-shot learning approaches.