A Kernel Redundancy Removing Policy for Convolutional Neural Network
share
Deep Convolutional Neural Networks (CNN) have won a significant place in the computer vision recently, which repeatedly convolving an image to extract the knowledge behind it. However, with the depth of convolutional layers getting deeper and deeper in recent years, the computational complexity also increases significantly, which make it difficult to be deployed on embedded systems with limited hardware resources. In this paper we propose a method to reduce the redundant convolution kernels during the computation of CNN and apply it to a network for super resolution (SR). Using PSNR drop compared to the original network as performance criterion, our method can get the optimal PSNR under a certain computation budget constraint. On the other hand, our method is also capable of minimizing the computation required under a given PSNR drop.
READ FULL TEXT
