Structured Pruning with Automatic Pruning Rate Derivation for Image Processing Neural Networks

Abstract

Structured pruning has been proposed for network model compression. Because most of existing structured pruning methods assign pruning rate manually, finding appropriate pruning rate to suppress the degradation of pruned model accuracy is difficult. Although we have been proposed the automatic pruning rate search method, the pruned model performances for complex image processing task such as ImageNet have not been evaluated. In this paper, we demonstrate a performance of the pruned model on ImageNet task using our proposed structured pruning method. Furthermore, we evaluate our pruning method in comparison of the pruned model performance using CIFAR-10 and ImageNet. When using ResNet-34 on ImageNet task, our proposed method reduces model parameters of ResNet-34 by 44.0% with 72.99% accuracy.