Browse Teach Curate

Mamba-Inspired Joint Unfolding Network for Snapshot Spectral Compressive Imaging

A deep unfolding network for reconstructing a 3D hyperspectral image from a single coded 2D measurement that reinterprets its accelerated half-quadratic-splitting stages through trapezoid discretization as a second-order ODE, restructures a Mamba selective state-space model into a Transformer-like global-to-local attention block, and integrates tensor mode-k unfolding to exploit low-rank structure across multiple scanning directions for sharper detail recovery.

Starscream Text

Detail Matters: Mamba-Inspired Joint Unfolding Network for Snapshot Spectral Compressive Imaging

In coded-aperture snapshot spectral imaging (CASSI), a 3D hyperspectral image is optically coded into a single 2D measurement and must be computationally reconstructed, an ill-posed inverse problem. …