ICLR 2026 · Conference paper

SpikeStereoNet: A Brain-Inspired Framework for Stereo Depth Estimation from Spike Streams

Zhuoheng Gao, Yihao Li, Jiyao Zhang, Rui Zhao, Tong Wu, Hao Tang, Zhaofei Yu, Hao Dong, Guozhang Chen, Tiejun Huang

SpikeStereoNet estimates dense stereo depth directly from microsecond spike camera streams using iterative recurrent spiking neural network refinement.

The scientific question

How can stereo depth be estimated directly from asynchronous spike streams?

Conventional frame cameras blur fast motion, whereas microsecond spike cameras lack dedicated stereo algorithms and annotated benchmarks for depth perception.

Synthetic and real spike stereo data training and evaluation pipeline.
Figure 1 from the original paper ↗

The core idea

Recurrent Spiking Stereo Refinement

SpikeStereoNet fuses raw binocular spike streams and iteratively refines disparity estimates through a recurrent spiking update module, evaluated on new synthetic and real-world stereo benchmarks.

What we found

Robust Depth in Challenging Scenes

The model outperforms prior approaches on textureless surfaces and extreme lighting while maintaining high depth accuracy when trained on substantially reduced data.

Continue the question

Related reading

Explore our research

Cite this work

Zhuoheng Gao, Yihao Li, Jiyao Zhang, Rui Zhao, Tong Wu, Hao Tang, Zhaofei Yu, Hao Dong, Guozhang Chen, Tiejun Huang. SpikeStereoNet: A Brain-Inspired Framework for Stereo Depth Estimation from Spike Streams. ICLR 2026.