TY - GEN
T1 - Truly shift-invariant convolutional neural networks
AU - Chaman, Anadi
AU - Dokmanić, Ivan
N1 - Publisher Copyright:
© 2021 IEEE
PY - 2021
Y1 - 2021
N2 - Thanks to the use of convolution and pooling layers, convolutional neural networks were for a long time thought to be shift-invariant. However, recent works have shown that the output of a CNN can change significantly with small shifts in input-a problem caused by the presence of downsampling (stride) layers. The existing solutions rely either on data augmentation or on anti-aliasing, both of which have limitations and neither of which enables perfect shift invariance. Additionally, the gains obtained from these methods do not extend to image patterns not seen during training. To address these challenges, we propose adaptive polyphase sampling (APS), a simple sub-sampling scheme that allows convolutional neural networks to achieve 100% consistency in classification performance under shifts, without any loss in accuracy. With APS, the networks exhibit perfect consistency to shifts even before training, making it the first approach that makes convolutional neural networks truly shift-invariant.
AB - Thanks to the use of convolution and pooling layers, convolutional neural networks were for a long time thought to be shift-invariant. However, recent works have shown that the output of a CNN can change significantly with small shifts in input-a problem caused by the presence of downsampling (stride) layers. The existing solutions rely either on data augmentation or on anti-aliasing, both of which have limitations and neither of which enables perfect shift invariance. Additionally, the gains obtained from these methods do not extend to image patterns not seen during training. To address these challenges, we propose adaptive polyphase sampling (APS), a simple sub-sampling scheme that allows convolutional neural networks to achieve 100% consistency in classification performance under shifts, without any loss in accuracy. With APS, the networks exhibit perfect consistency to shifts even before training, making it the first approach that makes convolutional neural networks truly shift-invariant.
UR - http://www.scopus.com/inward/record.url?scp=85108060102&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85108060102&partnerID=8YFLogxK
U2 - 10.1109/CVPR46437.2021.00377
DO - 10.1109/CVPR46437.2021.00377
M3 - Conference contribution
AN - SCOPUS:85108060102
T3 - Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
SP - 3772
EP - 3782
BT - Proceedings - 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2021
PB - IEEE Computer Society
T2 - 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2021
Y2 - 19 June 2021 through 25 June 2021
ER -