Skip to content
Book Open access

Hierarchical Temporal Residual Dynamics Network for Spiking Object Detectors

Nov 2026 · Proceedings of the 34th ACM International Conference on Multimedia · 0 citations · 39 references

Abstract

Spiking Neural Networks (SNNs) are promising for energy-efficient perception, yet their temporal advantage remains underexploited in high-precision object detection. In practice, existing directly trained SNN detectors often operate on replicated static frames or long-window integration, which suppresses temporal diversity, drives neurons toward homogeneous firing patterns, and limits the use of spike timing for fine-grained regression. This results in a fundamental mismatch: although SNNs are intrinsically temporal, their deployed dynamics for precise detection are largely absent. We present Hierarchical Temporal Residual Dynamics Network, a unified framework that reactivates temporal modeling at the neuron, feature, and decoding levels. Specifically, we introduce Temporal Residual Neurons to induce state-dependent firing dynamics under static inputs, Intra-/Inter-block Temporal Residual Connections to propagate temporal richness throughout the network, and a Dual-Path Spike Decoder to exploit fine-grained spike interval information for sub-rate bounding-box regression without extending the simulation horizon. Extensive experiments on three public benchmarks show that Hierarchical Temporal Residual Dynamics Network consistently surpasses strong baselines on both static and neuromorphic datasets with minimal overhead. Our findings suggest that hierarchical temporal residual modeling is key to unlocking the full potential of SNNs for high-precision perception.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.