TY - GEN
T1 - LST-Net: Learning a Convolutional Neural Network with a Learnable Sparse Transform
AU - Li, Lida
AU - Wang, Kun
AU - Li, Shuai
AU - Feng, Xiangchu
AU - Zhang, Lei
N1 - Publisher Copyright:
© 2020, Springer Nature Switzerland AG.
PY - 2020/8
Y1 - 2020/8
N2 - The 2D convolutional (Conv2d) layer is the fundamental element to a deep convolutional neural network (CNN). Despite the great success of CNN, the conventional Conv2d is still limited in effectively reducing the spatial and channel-wise redundancy of features. In this paper, we propose to mitigate this issue by learning a CNN with a learnable sparse transform (LST), which converts the input features into a more compact and sparser domain so that the spatial and channel-wise redundancy can be more effectively reduced. The proposed LST can be efficiently implemented with existing CNN modules, such as point-wise and depth-wise separable convolutions, and it is portable to existing CNN architectures for seamless training and inference. We further present a hybrid soft thresholding and ReLU (ST-ReLU) activation scheme, making the trained network, namely LST-Net, more robust to image corruptions at the inference stage. Extensive experiments on CIFAR-10/100, ImageNet, ImageNet-C and Places365-Standard datasets validated that the proposed LST-Net can obtain even higher accuracy than its counterpart networks with fewer parameters and less overhead.
AB - The 2D convolutional (Conv2d) layer is the fundamental element to a deep convolutional neural network (CNN). Despite the great success of CNN, the conventional Conv2d is still limited in effectively reducing the spatial and channel-wise redundancy of features. In this paper, we propose to mitigate this issue by learning a CNN with a learnable sparse transform (LST), which converts the input features into a more compact and sparser domain so that the spatial and channel-wise redundancy can be more effectively reduced. The proposed LST can be efficiently implemented with existing CNN modules, such as point-wise and depth-wise separable convolutions, and it is portable to existing CNN architectures for seamless training and inference. We further present a hybrid soft thresholding and ReLU (ST-ReLU) activation scheme, making the trained network, namely LST-Net, more robust to image corruptions at the inference stage. Extensive experiments on CIFAR-10/100, ImageNet, ImageNet-C and Places365-Standard datasets validated that the proposed LST-Net can obtain even higher accuracy than its counterpart networks with fewer parameters and less overhead.
UR - https://www.scopus.com/pages/publications/85097433490
U2 - 10.1007/978-3-030-58607-2_33
DO - 10.1007/978-3-030-58607-2_33
M3 - Conference article published in proceeding or book
AN - SCOPUS:85097433490
SN - 9783030586065
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 562
EP - 579
BT - Computer Vision – ECCV 2020 - 16th European Conference, 2020, Proceedings
A2 - Vedaldi, Andrea
A2 - Bischof, Horst
A2 - Brox, Thomas
A2 - Frahm, Jan-Michael
PB - Springer Science and Business Media Deutschland GmbH
T2 - 16th European Conference on Computer Vision, ECCV 2020
Y2 - 23 August 2020 through 28 August 2020
ER -