Learning Binary Sampling Patterns for Single-Pixel Imaging using Bilevel Optimisation
This work addresses a domain-specific problem for applications like single-pixel fluorescence microscopy, representing an incremental improvement in pattern optimization.
The paper tackles the problem of reconstructing objects in single-pixel imaging by learning task-specific binary illumination patterns using bilevel optimization, achieving superior reconstruction performance compared to baselines, particularly in highly undersampled scenarios.
Single-Pixel Imaging enables reconstructing objects using a single detector through sequential illuminations with structured light patterns. We propose a bilevel optimisation method for learning task-specific, binary illumination patterns, optimised for applications like single-pixel fluorescence microscopy. We address the non-differentiable nature of binary pattern optimisation using the Straight-Through Estimator and leveraging a Total Deep Variation regulariser in the bilevel formulation. We demonstrate our method on the CytoImageNet microscopy dataset and show that learned patterns achieve superior reconstruction performance compared to baseline methods, especially in highly undersampled regimes.