The authors encoded images using DINO, a self-supervised vision model, and trained a feature-conditioned decoder once to reconstruct images from those feature maps.
Small feature adapters remove visual domains while preserving scene content
A 2.9-million-parameter adapter modifies the small domain-specific component of frozen DINO features, enabling unpaired weather removal and synthetic-to-real translation.
Chinese Tech
Thomas Deixelberger · Markus Steinberger
Huawei · Graz University of Technology
Research Digest··3 min read
Deixelberger and Steinberger argue that conventional image translators preserve unwanted fog, rain or rendering style because their generators receive representations from which the source appearance can be recovered.
Why this paper
From Huawei and Graz University of Technology
In one line
A small adapter on frozen DINO features moves a 13-14% residue to translate domains and outperforms pixel-fed generators with fewer parameters.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§