metadata
pipeline_tag: video-classification
Event2Vec: Processing Neuromorphic Events Directly by Representations in Vector Space
Event2Vec is a novel representation that allows neural networks to process neuromorphic events directly. By drawing an analogy between words and events, this approach is fully compatible with the parallel processing capabilities of high-throughput Transformer architectures. It resolves the long-standing conflict between maintaining data sparsity and maximizing GPU efficiency, offering a promising balance for real-time, low-latency neuromorphic vision tasks.
- Paper: Event2Vec: Processing Neuromorphic Events Directly by Representations in Vector Space
- Repository: https://github.com/Intelligent-Computing-Lab-Panda/event2vec
Performance
The model demonstrates parameter efficiency, high throughput, and low latency across several benchmarks:
- DVS Gesture
- ASL-DVS
- DVS-Lip
It achieves high accuracy even with an extremely low number of events or low spatial resolutions.
Citation
If you find this work useful, please cite:
@misc{fang2025event2vecprocessingneuromorphicevents,
title={Event2Vec: Processing Neuromorphic Events Directly by Representations in Vector Space},
author={Wei Fang and Priyadarshini Panda},
year={2025},
eprint={2504.15371},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2504.15371},
}