https://therasr.github.io/

Thera: Aliasing-Free Arbitrary-Scale Super-Resolution with Neural
Heat Fields

Alexander Becker^[?],, Rodrigo Caye Daudt^[?],, Dominik Narnhofer^, 
Torben Peters^, Nando Metzger^, Jan Dirk Wegner^[?], Konrad
Schindler^
^[?]Equal contribution ^Photogrammetry and Remote Sensing, ETH Zurich
^[?]Department of Mathematical Modeling and Machine Learning,
University of Zurich
arXiv Code  Demo
Overview of the Thera method for aliasing-free super-resolution

Thera is the first arbitrary-scale super-resolution method with a
built-in physical observation model.

Show me some results!

First row: Comparison between input image and our super-resolved
result.
Second row: Comparison between a SOTA competitor (MSIT) and ours.
[input_crop]
[ours_cropp]
[msit_cropp]
[ours_cropp]
[input_crop]
[ours_cropp]
[msit_cropp]
[ours_cropp]
[input_crop]
[ours_cropp]
[msit_cropp]
[ours_cropp]
[input_crop]
[ours_cropp]
[msit_cropp]
[ours_cropp]
[input_crop]
[ours_cropp]
[msit_cropp]
[ours_cropp]
[input_crop]
[ours_cropp]
[msit_cropp]
[ours_cropp]
[input_crop]
[ours_cropp]
[msit_cropp]
[ours_cropp]
[input_crop]
[ours_cropp]
[msit_cropp]
[ours_cropp]
[input_crop]
[ours_cropp]
[msit_cropp]
[ours_cropp]
[input_crop]
[ours_cropp]
[msit_cropp]
[ours_cropp]

How it works

A hypernetwork estimates parameters $\{\mathbf{b}_1, \mathbf{W}_2\}^
{(i,j)}$ of pixel-wise, local neural heat fields. The phase shifts $\
mathbf{b}_1$ operate on globally learned components, before thermal
activations scale each component depending on their frequency and the
desired scaling factor. The components are then linearly combined
using coefficients $\mathbf{W}_2$, resulting in an
appropriately-blurred, continuous local neural field. This field is
then rasterized at the appropriate sampling rate (resolution) to
yield a part of the final output image (red square). Unlike previous
methods, correct anti-aliasing is guaranteed by design!

Thera method

Quantitative comparison

Due to its principled observation model, our method achieves
state-of-the-art performance on a variety of super-resolution
benchmarks.

Thera results scheme

Citation

If you found our work helpful, consider citing our paper :

@article{becker2025thera,
  title={Thera: Aliasing-Free Arbitrary-Scale Super-Resolution with Neural Heat Fields},
  author={Becker, Alexander and Daudt, Rodrigo Caye and Narnhofer, Dominik and Peters, Torben and Metzger, Nando and Wegner, Jan Dirk and Schindler, Konrad},
  journal={arXiv preprint arXiv:2311.17643},
  year={2025}
}

Website template based on Nerfies.