📝 SummarXiv

Abstract

Spike camera, a type of neuromorphic sensor with high-temporal resolution, shows great promise for high-speed visual tasks. Unlike traditional cameras, spike camera continuously accumulates photons and fires asynchronous spike streams. Due to unique data modality, spike streams require reconstruction methods to become perceptible to the human eye. However, lots of methods struggle to handle spike streams in low-light high-speed scenarios due to severe noise and sparse information. In this work, we propose Diff-SPK, the first diffusion-based reconstruction method for spike camera. Diff-SPK effectively leverages generative priors to supplement texture information in low-light conditions. Specifically, it first employs an \textbf{E}nhanced \textbf{T}exture \textbf{f}rom Inter-spike \textbf{I}nterval (ETFI) to aggregate sparse information from low-light spike streams. Then, ETFI serves as a conditioning input for ControlNet to generate the high-speed scenes. To improve the quality of results, we introduce an ETFI-based feature fusion module during the generation process. Moreover, we establish the first bona fide benchmark for the low-light spike stream reconstruction task. It significantly surpasses existing reconstruction datasets in scale and provides quantitative illumination information. The performance on real low-light spike streams demonstrates the superiority of Diff-SPK.