Style Transfer with Diffusion Models for Synthetic-to-Real Domain Adaptation
Estelle Chigot, Dennis G. Wilson, Meriem Ghrib, Thomas Oberlin · 2025
Read Style Transfer with Diffusion Models for Synthetic-to-Real Domain Adaptation by Estelle Chigot, Dennis G. Wilson, Meriem Ghrib, Thomas Oberlin (2025) 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-2505-16360.txt.
Description
Estelle Chigot, Dennis Wilson, Meriem Ghrib, and Thomas Oberlin use diffusion models for synthetic-to-real domain adaptation in a 2025 paper. Segmentation models trained on fake scenes often fail on real photos, most of all in bad weather, where labeled data is scarce. New foundation models can make lifelike images with no training at all.
The team restyle synthetic images toward the real look, class by class, so the labels stay true. One method tunes normalization per class with cross-attention maps. An extension filters those maps by feature likeness, to kill artifacts in odd spots. Segmentation trained on the restyled set then holds up far better on real scenes.
Catalog record
| CREATOR | Estelle Chigot, Dennis G. Wilson, Meriem Ghrib, Thomas Oberlin |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 48 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-16360