A Novel Weight-Shared Multi-Stage Network Architecture of CNNs for Scale Invariance

Convolutional neural networks (CNNs) have demonstrated remarkable results in image classification tasks for benchmark and practical uses. The CNNs with deeper architectures have achieved higher performances thanks to their numerous parameters and resulting high expression ability recently. However, the CNNs have a problem of limited robustness to geometric transformation of objects in images such as scaling and rotation. This problem is considered to limit performance improvement of the deep CNNs but there is no established solution. This study focuses on scale transformation and proposes a novel network architecture called weight-shared multi-stage network (WSMS-Net), which enables the existing deep CNNs, such as ResNet and DenseNet, to acquire robustness to scaling of objects. The WSMS-Net architecture consists of multiple stages of CNNs and is easily combined with existing deep CNNs. This study demonstrates that existing deep CNNs combined the proposed WSMS-Net archive higher accuracy for image classification tasks only with little increase in the number of parameters.