RobustSCI: Beyond Reconstruction to Restoration for Snapshot Compressive Imaging under Real-World Degradations
This work tackles the practical problem of video SCI performance degradation due to real-world imaging conditions, which is crucial for deploying SCI systems in unconstrained environments. It represents a significant step beyond current incremental improvements in SCI reconstruction.
This paper addresses the challenge of video Snapshot Compressive Imaging (SCI) under real-world degradations like motion blur and low light, shifting the focus from mere reconstruction to robust restoration of pristine scenes. The authors propose RobustSCI, a network incorporating a novel RobustCFormer block with deblur and frequency enhancement branches, and RobustSCI-C, which further integrates a post-processing deblurring network. Their methods significantly outperform existing SOTA models on new degraded testbeds and real-world data.
Deep learning algorithms for video Snapshot Compressive Imaging (SCI) have achieved great success, yet they predominantly focus on reconstructing from clean measurements. This overlooks a critical real-world challenge: the captured signal itself is often severely degraded by motion blur and low light. Consequently, existing models falter in practical applications. To break this limitation, we pioneer the first study on robust video SCI restoration, shifting the goal from "reconstruction" to "restoration"--recovering the underlying pristine scene from a degraded measurement. To facilitate this new task, we first construct a large-scale benchmark by simulating realistic, continuous degradations on the DAVIS 2017 dataset. Second, we propose RobustSCI, a network that enhances a strong encoder-decoder backbone with a novel RobustCFormer block. This block introduces two parallel branches--a multi-scale deblur branch and a frequency enhancement branch--to explicitly disentangle and remove degradations during the recovery process. Furthermore, we introduce RobustSCI-C (RobustSCI-Cascade), which integrates a pre-trained Lightweight Post-processing Deblurring Network to significantly boost restoration performance with minimal overhead. Extensive experiments demonstrate that our methods outperform all SOTA models on the new degraded testbeds, with additional validation on real-world degraded SCI data confirming their practical effectiveness, elevating SCI from merely reconstructing what is captured to restoring what truly happened.