Highly Efficient SNNs for High-speed Object Detection

TL;DR

This paper introduces a highly efficient Spiking Neural Network for object detection that achieves significant speedups and low latency on GPU and FPGA, addressing the high inference latency issue of traditional SNNs.

Contribution

The work presents a novel method to construct low-complexity, high-speed SNNs using a scale-aware pseudoquantization and a new continuous inference scheme with FewdIF neurons.

Findings

118X speedup on GPU with 1.5MB parameters

Achieves over 800 FPS on FPGA with low latency

Effective for real-time object detection

Abstract

The high biological properties and low energy consumption of Spiking Neural Networks (SNNs) have brought much attention in recent years. However, the converted SNNs generally need large time steps to achieve satisfactory performance, which will result in high inference latency and computational resources increase. In this work, we propose a highly efficient and fast SNN for object detection. First, we build an initial compact ANN by using quantization training method of convolution layer fold batch normalization layer and neural network modification. Second, we theoretically analyze how to obtain the low complexity SNN correctly. Then, we propose a scale-aware pseudoquantization scheme to guarantee the correctness of the compact ANN to SNN. Third, we propose a continuous inference scheme by using a Feed-Forward Integrate-and-Fire (FewdIF) neuron to realize high-speed object detection. Experimental results show that our efficient SNN can achieve 118X speedup on GPU with only 1.5MB parameters for object detection tasks. We further verify our SNN on FPGA platform and the proposed model can achieve 800+FPS object detection with extremely low latency.