Keyword [Stylized-ImageNet]
Geirhos R, Rubisch P, Michaelis C, et al. ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness[C]. international conference on learning representations, 2019.
1. Overview
In this paper, it shows that
1) ImageNet-trained CNNs are strongly biased towards recognising textures rather than shapes, which is contrast to human behavior.
2) CNN is able to learn a shape-based representation on Stylized-ImageNet (SIN), which fits human visual system, comes with benifits and becomes robust against distortions
2. Experiments
