PydNet: An Efficient CNN Architecture with Pyramid Depthwise Convolution Kernels

Convolutional neural networks (CNNs) have shown significant performance in solving various artificial intelligence tasks in recent years. However, the increasing model size has raised challenges in adopting them in limited-resource applications. Recently, many research works try to build efficient networks which are as small as possible and have small computation time while still have acceptable performance. A state-of-the-art architecture is ShuffleNetV2. It has Depthwise Separable Convolution (DWConvolution) in place of standard Convolution to reduce the model size. Its design is very efficient which follows many practical guild-lines. The main weakness of ShuffleNetV2 is small receptive field. This paper proposes an improved version ShuffleNetV2 which instead of using just a 3×3 kernel size for DWConvolution, it uses a pyramid kernel sizes to capture spatial information in many scales. The proposed architecture is evaluated on a highly competitive object recognition benchmark datasets (CIFAR-100). The experiments demonstrate that the proposed network achieves better performance compared with ShuffleNetV2 as well as other state-of-the-art networks in term of similar model size.