Dynamic Path-Controllable Deep Unfolding Network for Compressive Sensing

Area of Science:

• Computer Vision
• Signal Processing
• Machine Learning

Background:

• Deep Unfolding Networks (DUNs) excel in compressive sensing (CS) due to interpretability and performance.
• Current DUNs process all images through all stages, leading to computational inefficiency for simpler reconstructions.

Purpose of the Study:

• To propose a Dynamic Path-Controllable Deep Unfolding Network (DPC-DUN) for efficient CS reconstruction.
• To enable dynamic path selection for adaptive computational load and performance-complexity tradeoffs.

Main Methods:

• Developed a novel DPC-DUN architecture.
• Incorporated a path-controllable selector for dynamic route selection.
• Designed the network to be slimmable for adjustable performance-complexity.

Main Results:

• DPC-DUN demonstrates high flexibility in CS reconstruction.
• Achieved excellent performance with dynamic adjustment capabilities.
• Successfully addressed practical requirements for appealing CS solutions.

Conclusions:

• DPC-DUN offers a significant advancement in efficient and adaptable CS reconstruction.
• The dynamic path selection and slimmable nature provide practical advantages over traditional DUNs.