Beyond Single Solution: Multi-Hypothesis Collaborative Deep Unfolding Network for Image Compressive Sensing
Researchers have developed a new deep unfolding network called MHC-DUN for image compressive sensing. This network addresses the limitation of existing methods by considering multiple plausible solutions rather than a single one. It achieves this by jointly optimizing across diverse solution spaces, using dynamic step sizes for gradient descent and a refined proximal mapping module that considers correlations within and between hypotheses. A novel composite loss function ensures a balance between measurement fidelity, hypothesis diversity, and reconstruction accuracy, leading to superior performance compared to current CS networks. AI
IMPACT Introduces a novel approach to image compressive sensing by considering multiple hypotheses, potentially improving reconstruction quality and robustness.