TY - GEN
T1 - Learnable Polyphase Sampling for Shift Invariant and Equivariant Convolutional Networks
AU - Rojas-Gomez, Renan A.
AU - Lim, Teck Yian
AU - Schwing, Alexander G.
AU - Do, Minh N.
AU - Yeh, Raymond A.
N1 - Acknowledgments: We thank Greg Shakhnarovich & PALS at TTI-Chicago for the thoughtful discussions and computation resources. This work is supported in part by NSF under Grants 1718221, 2008387, 2045586, 2106825, MRI 1725729, NIFA award 2020-67021-32799, and funding by PPG Industries, Inc. We thank NVIDIA for providing a GPU.
We thank Greg Shakhnarovich & PALS at TTI-Chicago for the thoughtful discussions and computation resources. This work is supported in part by NSF under Grants 1718221, 2008387, 2045586, 2106825, MRI 1725729, NIFA award 2020-67021-32799, and funding by PPG Industries, Inc. We thank NVIDIA for providing a GPU.
PY - 2022
Y1 - 2022
N2 - We propose learnable polyphase sampling (LPS), a pair of learnable down/upsampling layers that enable truly shift-invariant and equivariant convolutional networks. LPS can be trained end-to-end from data and generalizes existing handcrafted downsampling layers. It is widely applicable as it can be integrated into any convolutional network by replacing down/upsampling layers. We evaluate LPS on image classification and semantic segmentation. Experiments show that LPS is on-par with or outperforms existing methods in both performance and shift consistency. For the first time, we achieve true shift-equivariance on semantic segmentation (PASCAL VOC), i.e., 100% shift consistency, outperforming baselines by an absolute 3.3%. Our project page and code are available at https://raymondyeh07.github.io/learnable_polyphase_sampling/.
AB - We propose learnable polyphase sampling (LPS), a pair of learnable down/upsampling layers that enable truly shift-invariant and equivariant convolutional networks. LPS can be trained end-to-end from data and generalizes existing handcrafted downsampling layers. It is widely applicable as it can be integrated into any convolutional network by replacing down/upsampling layers. We evaluate LPS on image classification and semantic segmentation. Experiments show that LPS is on-par with or outperforms existing methods in both performance and shift consistency. For the first time, we achieve true shift-equivariance on semantic segmentation (PASCAL VOC), i.e., 100% shift consistency, outperforming baselines by an absolute 3.3%. Our project page and code are available at https://raymondyeh07.github.io/learnable_polyphase_sampling/.
UR - http://www.scopus.com/inward/record.url?scp=85163167314&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85163167314&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85163167314
T3 - Advances in Neural Information Processing Systems
BT - Advances in Neural Information Processing Systems 35 - 36th Conference on Neural Information Processing Systems, NeurIPS 2022
A2 - Koyejo, S.
A2 - Mohamed, S.
A2 - Agarwal, A.
A2 - Belgrave, D.
A2 - Cho, K.
A2 - Oh, A.
PB - Neural information processing systems foundation
T2 - 36th Conference on Neural Information Processing Systems, NeurIPS 2022
Y2 - 28 November 2022 through 9 December 2022
ER -