Decomposing Private Image Generation via Coarse-to-Fine Wavelet Modeling
Jasmine Bayrooti, Weiwei Kong, Natalia Ponomareva, Carlos Esteves, Ameesh Makadia, Amanda Prorok · 2026
Read Decomposing Private Image Generation via Coarse-to-Fine Wavelet Modeling by Jasmine Bayrooti, Weiwei Kong, Natalia Ponomareva, Carlos Esteves, Ameesh Makadia, Amanda Prorok (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-2602-23262.txt.
Description
Jasmine Bayrooti and five co-authors propose a private route to image making based on coarse-to-fine wavelets. Models trained on sensitive images can recall and copy single cases, so strong privacy pledges matter. Standard private fine-tuning often wrecks image quality, above all in fine detail, because it adds noise to all weights.
The key idea is that the most private image content lives in coarse wavelet parts, such as facial traits and object shapes, while fine parts are largely generic and public. The method fine-tunes a spectral code model in private on low-detail wavelet numbers, then sharpens with a ready-made super-resolution model. Keeping the privacy budget to global structure yields better quality and style capture than rival private methods on MS-COCO and MM-CelebA-HQ. Later detail steps add no new leak, by a standard privacy rule.
Catalog record
| CREATOR | Jasmine Bayrooti, Weiwei Kong, Natalia Ponomareva, Carlos Esteves, Ameesh Makadia, Amanda Prorok |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-23262