EAS-SNN: Adaptive Sampling and Representation for Event-based Detection with Spiking Neural Networks
Core Concepts
Spiking Neural Networks offer a solution for adaptive event sampling in event-based detection, enhancing performance and energy efficiency.
Abstract
The content discusses the use of Spiking Neural Networks (SNNs) for adaptive event sampling in event-based detection. It introduces the concept of End-to-End Adaptive Sampling with Recurrent SNNs to address challenges in traditional dense neural networks. The study proposes a novel adaptive sampling module that leverages recurrent convolutional SNNs enhanced with temporal memory, achieving superior performance with fewer parameters and time steps. The approach surpasses existing state-of-the-art spike-based methods and demonstrates applicability beyond SNNs. Key highlights include:
Introduction to Event Cameras and their capabilities.
Challenges posed by asynchronous data captured by event cameras.
Integration of sampling-aggregation mechanisms in existing frameworks.
Introduction of Spiking Neural Networks (SNNs) for adaptive event sampling.
Proposal of an end-to-end learnable framework for event-based detection using SNNs.
Introduction of Residual Potential Dropout (RPD) and Spike-Aware Training (SAT).
Empirical evaluations on neuromorphic datasets showcasing superior performance.
EAS-SNN
Stats
4.4% mAP improvement on Gen1 dataset, requiring 38% fewer parameters and three time steps.
Quotes
"Spiking Neural Networks emerge as a natural fit for addressing the challenge of adaptive event sampling."
"Our approach demonstrably surpasses existing state-of-the-art spike-based methods."