Skip to main content

Research References

BICLab SpikeYOLO (ECCV 2024)​

Paper​

Title: "Integer-Valued Training and Spike-Driven Inference Spiking Neural Network for High-performance and Energy-efficient Object Detection"

Key Contributions​

  • Integer-valued training for spiking neural networks
  • Spike-driven inference for energy efficiency
  • Adaptation of YOLO architecture for SNNs
  • I-LIF (Integer-valued LIF) neuron implementation

Repository​

The original BICLab SpikeYOLO implementation serves as the foundation for this project.

eTraM Dataset​

Dataset Information​

Name: Event-based Traffic Monitoring Dataset

Characteristics:

  • Resolution: 1280×720 pixels
  • Classes: 8 traffic participant classes
  • Format: HDF5 event files with NumPy annotations
  • Camera: Prophesee EVK4 HD (Sony IMX636 sensor)

Classes​

  1. Pedestrian
  2. Car
  3. Bicycle
  4. Bus
  5. Motorbike
  6. Truck
  7. Tram
  8. Wheelchair

ByteTracker​

Tracking Algorithm​

ByteTracker is used for object tracking in this project:

  • Algorithm: Hungarian algorithm-based tracking
  • Features: Handles occlusions and re-identifications
  • Association: Uses detection features for object association

Event-Based Vision​

Event cameras capture changes in brightness at each pixel independently, providing:

  • High temporal resolution (microsecond precision)
  • High dynamic range (>86 dB)
  • Low latency (event-by-event processing)
  • Energy efficiency (only processes changes)

Spiking Neural Networks​

SNNs are biologically-inspired neural networks that:

  • Process information through spikes
  • Operate on discrete time steps
  • Offer energy efficiency benefits
  • Enable temporal processing

Object Detection​

YOLO (You Only Look Once) architecture adapted for:

  • Event-based input
  • Spiking neural network processing
  • Real-time object detection
  • Multi-class detection

Additional Resources​

Event Camera Resources​

Spiking Neural Networks​

  • SNN Research: Active research area in neuromorphic computing
  • Energy Efficiency: Key advantage of SNNs for edge devices

Object Detection​

  • YOLO: Popular real-time object detection framework
  • Detection Metrics: mAP, precision, recall for evaluation

Acknowledgments​

This project builds upon:

  • BICLab SpikeYOLO implementation
  • eTraM dataset
  • ByteTracker algorithm
  • YOLO architecture

License​

Please refer to the repository for license information.