Learning spatially adaptive sparsity level maps for arbitrary convolutional dictionaries
Joshua Schulz, David Schote, Christoph Kolbitsch, Kostas Papafitsoros, Andreas Kofler · 2026
Read Learning spatially adaptive sparsity level maps for arbitrary convolutional dictionaries by Joshua Schulz, David Schote, Christoph Kolbitsch, Kostas Papafitsoros, Andreas Kofler (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-21707.txt.
Description
Learned image reconstruction often leans on black-box modules whose inner logic is hard to read and whose robustness is in doubt. Joshua Schulz and four co-authors build on a method that embeds data-driven detail into a model-based convolutional dictionary through neural-network-inferred maps of local sparsity levels. Better network design and training extend the method so filter order no longer matters and the dictionary can be swapped at run time.
Tests on low-field MRI, including live human data, show the gain from switching dictionaries, and the method beats other recent deep learning methods there. On out-of-distribution data it degrades less than rivals, which the authors credit to its model-based core leaning less on training data.
Catalog record
| CREATOR | Joshua Schulz, David Schote, Christoph Kolbitsch, Kostas Papafitsoros, Andreas Kofler |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-21707