Revisiting Dynamic Convolution via Matrix Decomposition
arXiv:2103.08756
Abstract
Recent research in dynamic convolution shows substantial performance boost for efficient CNNs, due to the adaptive aggregation of K static convolution kernels. It has two limitations: (a) it increases the number of convolutional weights by K-times, and (b) the joint optimization of dynamic attention and static convolution kernels is challenging. In this paper, we revisit it from a new perspective of matrix decomposition and reveal the key issue is that dynamic convolution applies dynamic attention over channel groups after projecting into a higher dimensional latent space. To address this issue, we propose dynamic channel fusion to replace dynamic attention over channel groups. Dynamic channel fusion not only enables significant dimension reduction of the latent space, but also mitigates the joint optimization difficulty. As a result, our method is easier to train and requires significantly fewer parameters without sacrificing accuracy. Source code is at https://github.com/liyunsheng13/dcd.
Accepted by ICLR 2021
References in corpus (8)
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- Exploiting Linear Structure Within Convolutional Networks for Efficient Evaluation
- Speeding-up Convolutional Neural Networks Using Fine-tuned CP-Decomposition
- CondConv: Conditionally Parameterized Convolutions for Efficient Inference
- Rethinking Bottleneck Structure for Efficient Mobile Network Design
- Dynamic Region-Aware Convolution
- Dynamic Group Convolution for Accelerating Convolutional Neural Networks
- WeightNet: Revisiting the Design Space of Weight Networks