The authors introduce Pairwise Artifacts Learning (PAL), a pretraining stage that contrasts edit-related and non-related image pairs using a pairwise binary cross-entropy objective.
Pairwise edit training helps localize manipulated image regions more reliably
The authors train forensic encoders to distinguish edit-related image pairs before adapting them to pixel-level manipulation localization.
Chinese Tech
Xuekang Zhu · Kaiwen Feng · Ruifeng Wang · Xiwen Wang · Xiaochen Ma · Bo Du · +7 more
Sichuan University · Ant Group · The Hong Kong University of Science and Technology · Wuhan University · South China University of Technology
Research Digest··2 min read
Zhu and colleagues recast image manipulation localization as a latent-variable problem, arguing that models must infer subtle editing artifacts before converting them into segmentation masks.
Why this paper
From Ant Group and 6 others
In one line
Modeling artifacts explicitly via pairwise edit relations improves image manipulation localization.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·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.
§