Investigating and Explaining the Frequency Bias in Image Classification
This addresses a specific phenomenon in CNN behavior for image classification researchers, but it is incremental as it builds on known differences from human vision.
The paper investigates the frequency bias in CNNs for image classification, showing that high-frequency components are less utilized than low- and mid-frequency ones, and identifies spectral density and class consistency as key factors affecting this bias.
CNNs exhibit many behaviors different from humans, one of which is the capability of employing high-frequency components. This paper discusses the frequency bias phenomenon in image classification tasks: the high-frequency components are actually much less exploited than the low- and mid-frequency components. We first investigate the frequency bias phenomenon by presenting two observations on feature discrimination and learning priority. Furthermore, we hypothesize that (i) the spectral density, (ii) class consistency directly affect the frequency bias. Specifically, our investigations verify that the spectral density of datasets mainly affects the learning priority, while the class consistency mainly affects the feature discrimination.