AI Paper Atlas

Research map / Detection and segmentation / CNN features

Representative Forgery Mining for Fake Face Detection

Chengrui Wang, Weihong Deng

CVPR 2021 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 266 citations (OpenAlex)

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Add RFM to XceptionXception · DFFDTDR at FDR 0.01%: 77.92 -> 90.44Improved
Add RFM to XceptionXception · Celeb-DFTDR at FDR 0.1%: 89.11 -> 93.88Improved
Add RFM to PatchPatch · DFFD Group ATDR at FDR 0.01%: 99.23 -> 99.68Improved
Add RFM to PatchPatch · DFFD Group BTDR at FDR 0.1%: 99.58 -> 99.56No clear effect
Replace baseline training with RFM trainingXception · DFFD Group BTDR at FDR 0.01%: 83.46 -> 93.80Improved
Use FAM-guided multiple erasing blocks rather than the baselineXception · DFFDTDR at FDR 0.01%: +8.33%Improved
Add multiple erasing blocks with random anchors (ablation)RFM · DFFDTDR at FDR 0.01%: +0.51%Improved
Add FAM guidance for selecting sensitive regions (ablation)RFM · DFFDTDR at FDR 0.01%: +2.21%Improved
Replace SFE with progressive erasing (PSFE) (ablation)RFM · DFFDTDR at FDR 0.01%: 92.84 vs 92.78; 95.34 vs 95.50No clear effect
Increase erase size from 30 to 120 at p 0.5 (ablation)RFM · Celeb-DF-subsetAAUC: 97.89 -> 96.79Hurt
Increase erase probability from 0.5 to 1.0 at size 30 (ablation)RFM · Celeb-DF-subsetAAUC: 97.89 -> 97.32Hurt
Use Adversarial Erasing instead of RFM (ablation)Xception · Celeb-DFTDR at FDR 0.01%: 84.22 -> 76.63Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy