AI Paper Atlas

Research map / Detection and segmentation / CNN features

Exploring Specular Reflection Inconsistency for Generalizable Face Forgery Detection

Hongyan Fei, Zexi Jia, Chuanwei Huang, Jinchao Zhang, Jie Zhou

ICLR 2026 · Detection and segmentation › CNN features · approach: Forgery detection · 3 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 specular reflection branch to image branch (ablation)SRI-Net · DiFFAUC (%): 66.0 -> 83.9Improved
add MSR texture and direct light with two-stage cross-attention to Img+SPR (ablation)SRI-Net · DiFFAUC (%): 83.9 -> 86.8Improved
replace BFM mean texture with Retinex-based T_msr (ablation)SRI-Net · DiFFAUC (%): 79.9 -> 86.8Improved
replace BFM PCA-fitted texture with Retinex-based T_msr (ablation)SRI-Net · DF40AUC (%): 88.2 -> 90.9Improved
add shape normalization (ablation)SRI-Net · DF40AUC (%): 83.7 -> 90.9Improved
use cross-attention instead of SE-block fusion (ablation)SRI-Net · DiFFAUC (%): 84.5 -> 86.8Improved
use cross-attention instead of concatenation (ablation)SRI-Net · CelebDF-v2AUC (%): 85.6 -> 87.5Improved
apply Gaussian blurring to test images (ablation)SRI-Net · DF40AUC (%): 90.9 -> 88.7Hurt
apply JPEG compression to test images (ablation)SRI-Net · DF40AUC (%): 90.9 -> 90.2No clear effect
apply Gaussian noise to test images (ablation)SRI-Net · DiFFAUC (%): 86.8 -> 85.5Hurt
use MSR texture alone as input (ablation)SRI-Net · DiFFAUC (%): 66.0 (Img) vs 60.8 (MSR)Hurt

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