Efficient Deep Demosaicing with Spatially Downsampled Isotropic Networks
Cory Fan, Wenchao Zhang · 2026
Read Efficient Deep Demosaicing with Spatially Downsampled Isotropic Networks by Cory Fan, Wenchao Zhang (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-2601-00703.txt.
Description
Cory Fan and Wenchao Zhang revisit how camera images are reconstructed from sensor data. A camera sensor records only one color per pixel through a color filter array, and demosaicing rebuilds the full color image, a first step in every photo pipeline. Deep networks do this well, but the residual-style networks usually used avoid shrinking the image spatially, making them too expensive for phones.
The authors argue the opposite choice pays off: downsampling the image substantially inside an isotropic network can raise both efficiency and quality. They design simple fully convolutional networks to test that claim on demosaicing and on the joint task of demosaicing with denoising. If it holds, high-quality reconstruction becomes far more practical on mobile hardware.
Catalog record
| CREATOR | Cory Fan, Wenchao Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-00703