A novel mechanism for converting convolutional neural networks (CNNs) to spiking neural networks (SNNs) to facilitate ready deployment, that is, mapping on spiking hardware architectures, is proposed in this paper.

The authors have provided a detailed analysis of CNNs and their applications whilst highlighting the deep learning nature of such neural networks. Fundamentally, CNNs have been mapped on conventional central processing units (CPUs) with numerical processing capabilities. However, modern-day versions of CNNs with increasing complexities demand high-performance processing capability, which essentially limits their ready adoption.

Without incurring significant performance loss, the scheme proposed by the authors converts the architecture type from CNN to SNN. The assessment of the proposed scheme is conducted on the Neovision2 tower video dataset and the CIFAR-10 image dataset. The resulting improvement in performance in terms of power consumption on off-the-shelf hardware such as field-programmable gate arrays (FPGAs) was quantified and elaborated upon.

The scheme presented in the paper alongside the analysis will be of use to the low-power computer vision community. The readership is expected to have an understanding of neural networks and computer vision to comprehend the proposed scheme.