Latent Posterior-Mean Rectified Flow for Higher-Fidelity Perceptual Face Restoration
Xin Luo, Menglin Zhang, Yunwei Lan, Tianyu Zhang, Rui Li, Chang Liu, Dong Liu · 2025
Read Latent Posterior-Mean Rectified Flow for Higher-Fidelity Perceptual Face Restoration by Xin Luo, Menglin Zhang, Yunwei Lan, Tianyu Zhang, Rui Li, Chang Liu, Dong Liu (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-2507-00447.txt.
Description
Restoring a damaged face photo forces a trade: sharpen it and the likeness drifts, stay faithful and it looks fake. Xin Luo and six co-authors work that trade in 2025 by moving a recent flow-based restoration method into the compressed latent space of an autoencoder.
Working in latent space aligns the repair with how humans judge pictures, rather than raw pixels. The best achievable faithfulness is then capped by the autoencoder's own rebuilding error, so the team designed a stronger autoencoder for the job. On blind face restoration tests their method beat rivals on the perception-faithfulness balance and converged nearly six times faster than the pixel-space version. Code is promised as open source.
Catalog record
| CREATOR | Xin Luo, Menglin Zhang, Yunwei Lan, Tianyu Zhang, Rui Li, Chang Liu, Dong Liu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-00447