Dual-domain self-supervised feature alignment via spectral–spatial representation learning for deepfake anomaly detection
Yi He · 2026
Read Dual-domain self-supervised feature alignment via spectral–spatial representation learning for deepfake anomaly detection by Yi He (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/plos-pone-0358049.txt.
Description
Yi He published a 2026 methods paper in PLOS ONE on spotting deepfakes, manipulated images and videos that fool viewers, without needing labeled fakes for training. Most detectors learn from collections of marked forgeries and still miss subtle edits. He's framework, SDDA, trains itself instead, comparing each image in two domains at once: its ordinary appearance and its frequency pattern, which records fine repeating detail.
Two encoders extract features from both views and an alignment step forces them to agree, while four custom measures flag inconsistencies in the learned features. A contrastive objective sharpens the separation between normal and odd patterns, and the author reports the system stayed robust under compression and lighting changes that often defeat detectors. In the reported experiments it beat leading baselines, especially on manipulation types it had never seen. The design aims at an interpretable detector that does not depend on huge labeled fake collections.
Catalog record
| CREATOR | Yi He |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63543 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |