A Simple Adaptive Unfolding Network for Hyperspectral Image Reconstruction

  • 2023-01-24 18:28:21
  • Junyu Wang, Shijie Wang, Wenyu Liu, Zengqiang Zheng, Xinggang Wang
  • 0


We present a simple, efficient, and scalable unfolding network, SAUNet, tosimplify the network design with an adaptive alternate optimization frameworkfor hyperspectral image (HSI) reconstruction. SAUNet customizes a ResidualAdaptive ADMM Framework (R2ADMM) to connect each stage of the network via agroup of learnable parameters to promote the usage of mask prior, which greatlystabilizes training and solves the accuracy degradation issue. Additionally, weintroduce a simple convolutional modulation block (CMB), which leads toefficient training, easy scale-up, and less computation. Coupling these twodesigns, SAUNet can be scaled to non-trivial 13 stages with continuousimprovement. Without bells and whistles, SAUNet improves both performance andspeed compared with the previous state-of-the-art counterparts, which makes itfeasible for practical high-resolution HSI reconstruction scenarios. We set newrecords on CAVE and KAIST HSI reconstruction benchmarks. Code and models areavailable at https://github.com/hustvl/SAUNet.


Quick Read (beta)

loading the full paper ...