CSGN: Combined Channel- and Spatial-Wise Dynamic Gating Architecture for Convolutional Neural Networks

The explosive computation and memory requirements of convolutional neural networks (CNNs) hinder their deployment in resource-constrained devices. Because conventional CNNs perform identical parallelized computations even on redundant pixels, the saliency various features an image should be reflected for higher energy efficiency market penetration. This paper proposes a novel channel spatial gating network (CSGN) adaptively selecting vital channels generating spatial-wise execution masks. A CSGN can characterized as dynamic spatial-aware module by maximally utilizing opportunistic sparsity. Extensive experiments were conducted CIFAR-10 ImageNet datasets based ResNet. results revealed that, with proposed architecture, amount multiply-accumulate (MAC) operations was reduced 1.97–11.78× 1.37–13.12× ImageNet, respectively, negligible accuracy degradation inference stage compared baseline architectures.