Learning Filter Basis for Convolutional Neural Network Compression

By Yawei Li, Shuhang Gu, Luc Van Gool, Radu Timofte. 2019

106 citations recorded by the source corpus.

Convolutional neural networks (CNNs) based solutions have achieved state-of-the-art performances for many computer vision tasks, including classification and super-resolution of images. Usually the success of these methods comes with a cost of millions of parameters due to stacking deep convolutional layers. Moreover, quite a large number of filters are also used for a single convolutional layer, which exaggerates the parameter burden of current methods. Thus, in this paper, we try to reduce the number of parameters of CNNs by learning a basis of the filters in convolutional layers. For the forward pass, the learned basis is used to approximate the original filters and then used as parameters for the convolutional layers. We validate our proposed solution for multiple CNN architectures on image classification and image super-resolution benchmarks and compare favorably to the existing state-of-the-art in terms of reduction of parameters and preservation of accuracy. Code is available at https://github.com/ofsoundof/learning_filter_basis

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Reparameterizing Convolutions for Incremental Multi-Task Learning without Task Interference (2020, background): “In particular, filter-wise decomposition into a product of low-rank filters [20], filter groups [47], a basis of filter groups [30], etc.”
Blending Anti-Aliasing into Vision Transformer (2021, methodology): “The linear assembling strategy has been proven effective in computational photography [43], parametric human body representation [44], network compression [45] and acceleration [46].”
Provable Filter Pruning for Efficient Neural Networks (2019, background): “Filter Thresholding (FT, Li et al. (2016) Consider the set of filtersW ` = [W ` 1 , .”
T-Basis: a Compact Representation for Neural Networks (2020, background): “Han et al. (2015; 2016) proposed to prune ‘low impact’ neurons, while other approaches focused more on pruning filters/channels (Li et al., 2017; Gordon et al., 2018; Yang et al., 2018), as the latter leads to more regular kernel shapes (Li et al., 2019b).”