2026 BrainChip Discussion, page-73

7.3 KAUST-Neuromorphic Traffic Scenes Datasets

Existing event-based datasets for event-based object detection remain limited in number, geographic scope, and class breadth. KAUST-NTS attempted to address these gaps as the first Middle East–based neuromorphic traffic dataset, emphasizing challenging high-illumination conditions and four key traffic classes.125 Even in its unfinished state, it comprised approximately 60,000 labeled instances collected from multiple 60-second recordings and additionally included Instance Segmentation per-pixel masks. The whole project was presented in Chapter 5.

7.4 Full Neuromorphic System Implementation

Chapter 6 expanded the scope of this research to the limited adoption of neuromorphic processors for the target task by presenting a case study of an end-to-end neuromorphic system. A complete low-power implementation was demonstrated using IniVation’s DVXplorer Lite event-based camera and BrainChip’s Akida ADK1000 spiking platform.

Evaluation across three datasets (real-world and synthetic) yielded promising results:

• Achieved a 27.1 mAP@50 on GEN1 dataset (49.6% of state-of-the-art performance)
• Reached 83% of the best-reported performance on PeDRo dataset at the time with 17 times fewer parameters
• Demonstrated excellent edge computing characteristics:– Power consumption: 66mW– Processing latency: 138.88ms

These results validate the system’s suitability for real-time edge applications, offering an excellent balance between performance and resource efficiency.

7.5 Final Remarks

This dissertation presented six significant contributions to the field of event-based object detection:

1. ReYOLOv8: A baseline framework for spatiotemporal processing that achieves high performance with compact architectures suitable for real-time operation126
2. VTEI: A lightweight event encoding method that balances performance with edge computing constraints
3. RPS:Anefficientdataaugmentation technique that leverages event-specific characteristics to improve model performance
4. Chimera: A framework for efficiently exploring design spaces for eventbased detectors, producing architectures optimized for real-world deployment
5. KAUST-NTS dataset: First Middle East-based dataset for neuromorphic traffic object detection
6. Neuromorphic System Implementation: A complete end-to-end neuromorphic solution combining event-based camera and spiking neural network hardware, demonstrating practical viability for edge applications with minimal power consumption

Collectively, these contributions advance the state of the art in neuromorphic computing and provide a comprehensive solution to the challenge of designing efficient and effective event-based object detection systems. The methods and frameworks developed in this research offer promising avenues for future work in neuromorphic vision systems and their application in time-critical domains such as autonomous vehicles, robotics, and human-machine interaction.

https://repository.kaust.edu.sa/server/api/core/bitstreams/db0977b1-959d-4d32-971e-40dc212ba068/content

I posted a link to an abridged version of this report the above extract is taken from the full article the link for which is now provided.

My opinion only DYOR

Fact Finder