-
style bias vs shape bias카테고리 없음 2022. 7. 5. 17:06
Reducing domain gap by reducing style bias : cvpr2022
What shapes feature representations? Exploring datasets, architectures, and training 2020 neurlps
https://proceedings.neurips.cc/paper/2020/file/71e9c6620d381d60196ebe694840aaaa-Paper.pdf
Texture-like representation of objects in human visual cortex
: 2022 03, 이상한곳
https://www.pnas.org/doi/abs/10.1073/pnas.2115302119
A Developmentally-Inspired Examination of Shape versus Texture Bias in Machines : 2022 05, arXiv
https://arxiv.org/abs/2202.08340
encoding 이후 cos-sim끼리 비교, anchor가 texture match랑 가까우면 texture bias
The origins and prevalence of texture bias in convolutional neural networks : 2020 Neurlps
https://arxiv.org/pdf/1911.09071.pdf
The Role of Shape for Domain Generalization on Sparsely-Textured Images(CVPR 2022)Rethinking the image feature biases exhibited by deep convolutional neural network models in image recognition : 2022, CAAI
https://ietresearch.onlinelibrary.wiley.com/doi/full/10.1049/cit2.12097
Decomposing Texture and Semantics for Out-of-distribution Detection ICLR 2022. https://openreview.net/forum?id=UYDtmk6BMf5
texture : DFT
semantic : ssvd로 구함
오픈리뷰에서 둘 사이 여전히 entangled되어있다고 까임
A Closer Look at Fourier Spectrum Discrepancies for CNN-Generated Images Detection (CVPR 2021)
Think Twice Before Detecting GAN-generated Fake Images from their Spectral Domain Imprints(CVPR 2022)
Spatial Frequency Bias in Convolutional Generative Adversarial Networks(AAAI 2022)
file:///Users/dongkyunkim/Downloads/20675-Article%20Text-24688-1-2-20220628.pdf
Baseline
ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness(ICLR2019)