Reproduce “Shared-kernel Wavelet Neural Networks for Poisson Image Reconstruction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuanhao Gong, Tan Tang, Qianyan Liu
A compact neural network, the Shared-kernel Wavelet Neural Network (SKWNN), reconstructs images from their sparse Laplacian fields with high accuracy. This architecture, featuring only 177 parameters, offers efficient and resolution-agnostic image reconstruction, outperforming traditional wavelet methods.
Key insights
- Laplacian fields provide a sparse, statistically stable, and uniquely reconstructible image representation, offering benefits for compression and processing.
- A deep learning model can effectively learn optimal multiscale filters, similar to wavelets, to perform highly accurate Poisson image reconstruction.
- Employing shared convolution kernels across scales in a wavelet-inspired neural network architecture drastically reduces model complexity while ensuring generalization to arbitrary image resolutions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.