AI Paper Atlas

Research map / Detection and segmentation / CNN features

ResProto-FD: Visual-Language Residual Prototype Sets for Generalized Face Forgery Detection

Jiuyao Jing, Yu Zheng, Chunlei Peng

AAAI 2026 · Detection and segmentation › CNN features · approach: Forgery detection · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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 VLRL to CLIP baseline (ablation)CLIP ViT-B/16 detector · DF40 average (six methods)AUC: 86.0 -> 88.9Improved
add VLRL to CLIP baseline (ablation)CLIP ViT-B/16 detector · CDF+DFD+DFDC averageAUC: 83.1 -> 83.2No clear effect
add GRP alone to CLIP baseline (ablation)CLIP ViT-B/16 detector · CDF+DFD+DFDC averageAUC: 83.1 -> 83.1No clear effect
add GRP on top of VLRL (ablation)ResProto-FD · CDF+DFD+DFDC averageAUC: 83.2 -> 84.9Improved
add GRP on top of VLRL (ablation)ResProto-FD · DF40 average (six methods)AUC: 88.9 -> 90.5Improved
use residual regularization L_RL without residual features f_res (ablation)VLRL · CDF+DFD+DFDC averageAUC: 83.1 -> 79.0Hurt
use residual features f_res alone without L_RL (ablation)VLRL · CDF+DFD+DFDC averageAUC: 83.1 -> 82.9No clear effect
combine f_res with L_RL (full VLRL) (ablation)VLRL · DF40 average (six methods)AUC: 86.0 -> 88.9Improved
add score decay (gamma 0.99 vs 1.0) to GRP (ablation)GRP · DF40 average (six methods)AUC: 90.1 -> 90.5Improved
vary top-K prototypes (16, 32, 64, 128) (ablation)GRP inference · CDF, DFD, DFDC, DF40 subsetsAUCMixed

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