A naukri.com initiative
Arxiv
2w
258
Image Credit: Arxiv
SpikeVideoFormer: An Efficient Spike-Driven Video Transformer with Hamming Attention and $\mathcal{O}(T)$ Complexity
- SpikeVideoFormer is introduced as an efficient spike-driven video Transformer with linear temporal complexity O(T).
- The model features a spike-driven Hamming attention (SDHA) that transitions from traditional real-valued attention to spike-driven attention.
- Multiple spike-driven space-time attention designs were analyzed to identify an optimal scheme for video tasks with linear temporal complexity.
- The SpikeVideoFormer model demonstrates superior performance in diverse video tasks like classification, human pose tracking, and semantic segmentation, outperforming existing SNN approaches and offering significant efficiency gains.
Read Full Article
15 Likes
For uninterrupted reading, download the app