The authors encoded degraded images using a frozen DINOv3-L vision transformer and averaged features from 23 layers.
Fused DINOv3 features enable fast, faithful real-world image super-resolution
RAESR restores images in a frozen vision model’s latent space, outperforming a matched VAE-based system and improving benchmark trade-offs between fidelity, perceptual quality and speed.
Chinese Tech
Wanzhou Lei · Cuifeng Sheng · Yanjin He · Maohua Li · Hua Yuan · Per-Olof Persson · +1 more
University of California, Berkeley · Alibaba Group · University of Michigan, Ann Arbor · Nanjing University · Southeast University
Research Digest··3 min read
Lei and colleagues treat super-resolution as a mapping from degraded image features back toward the features of clean, natural images.
Why this paper
From Alibaba Group and 4 others
In one line
Super-resolution achieves best fidelity-perception trade-off when performed in the frozen fused latent space of DINOv3-L.
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.
§