Core Concepts
A specialized spiking feature pyramid network (SpikeFPN) optimized for efficient automotive event-based object detection, leveraging the unique temporal dynamics and sparse computation capabilities of spiking neurons.
Abstract
The paper explores the use of Spiking Neural Networks (SNNs) for automotive object detection, which is a challenging task due to the sparse and asynchronous nature of event-based sensor data. The authors propose a specialized spiking feature pyramid network (SpikeFPN) that effectively leverages the temporal dynamics and sparse computation capabilities of spiking neurons.
Key highlights:
- The authors investigate the membrane potential dynamics of spiking neurons and introduce an adaptive threshold mechanism to enhance the stability and efficiency of the training process.
- The SpikeFPN architecture is designed to capture the unique temporal characteristics of spiking neurons, with a multi-stage downsampling backbone and a spiking feature pyramid module.
- Comprehensive evaluations on the GEN1 Automotive Detection (GAD) benchmark dataset demonstrate that SpikeFPN outperforms both traditional SNNs and advanced Artificial Neural Networks (ANNs) with attention mechanisms, achieving a mean Average Precision (mAP) of 0.477.
- The efficient design of SpikeFPN, leveraging the inherent sparse computation capabilities of spiking neurons, ensures robust performance while optimizing computational resources.
- Ablation studies and experiments on the N-CARS dataset further validate the effectiveness and generalizability of the proposed approach.
Stats
The paper reports the following key metrics:
mAP50: 0.477
mAP50:95: 0.223
Inference speed: 54 frames per second on a Tesla V100 GPU
Quotes
"Evidently, SpikeFPN achieves a mean Average Precision (mAP) of 0.477 on the GEN1 Automotive Detection (GAD) benchmark dataset, marking significant increases over the selected SNN baselines."
"Moreover, the efficient design of SpikeFPN ensures robust performance while optimizing computational resources, attributed to its innate sparse computation capabilities."