SGF-CDNet: A Consistency-Discrepancy Graph Network over Semantic-Geometric Fused Nodes for Face Forgery Detection

Authors: Jiayao Jiang, Bin Liu, Nenghai Yu

Published: 2026-07-04 13:59:59+00:00

Comment: 6 pages, 4 figures. This work has been accepted at ICMEW 2026

AI Summary

SGF-CDNet proposes a novel approach for face forgery detection by identifying subtle disharmony among different facial regions. It utilizes a Consistency-Discrepancy Graph Network (CD-GNN) built upon Semantic-Geometric Fused (SGF) nodes, which combine semantic regions from face parsing with geometric information from facial landmarks. This dual-path GNN concurrently analyzes consistency and discrepancy within facial components, effectively uncovering structural tensions introduced by forgeries.

Abstract

The rapid advancement of deepfakes necessitates robust face forgery detection. Although forged faces may lack obvious artifacts, they often contain subtle disharmony among different facial regions. We propose SGF-CDNet, a Consistency-Discrepancy Graph Network (CD-GNN) over Semantic-Geometric Fused (SGF) nodes. First, SGF-CDNet constructs SGF nodes by deeply fusing semantic regions from face parsing with geometric information from facial landmarks, allowing nodes to capture both high-level concepts and precise geometric constraints. Next, a dual-path CD-GNN performs parallel relational reasoning on these nodes across two dimensions: consistency and discrepancy. The consistency path evaluates if facial components follow natural biological patterns, while the discrepancy path mines for structural tensions and feature conflicts introduced by forgeries. By integrating these processes, our model effectively identifies disharmonious relationships between facial components. Extensive experiments on public datasets demonstrate that SGF-CDNet achieves superior performance, establishing it as a reliable solution for face forgery detection.

Key findings

SGF-CDNet achieved superior performance and generalization capabilities across various unseen datasets, outperforming 11 representative baselines. Ablation studies confirmed the effectiveness of SGF nodes, the dual-path CD-GNN, and the gating mechanism. The model demonstrated robustness to various real-world degradations like compression, occlusion, blur, moderate noise, head pose variations, and moderate preprocessing perturbations.

Approach

The SGF-CDNet first constructs SGF nodes by fusing semantic regions (from face parsing) and geometric information (from facial landmarks). Then, a dual-path CD-GNN performs relational reasoning: a consistency path checks natural biological patterns, and a discrepancy path identifies structural tensions and feature conflicts. These reasoned nodes are used to enhance the image encoder's feature maps via a gating mechanism.

Datasets

FaceForensics++ (FF++), Celeb-DeepFake-v2 (CDF2), DeepFake Detection Challenge (DFDC), DeepFake Detection Challenge Preview (DFDCP), FFIW-10K (FFIW)

Model(s)

ConvNeXt-Base (as image encoder), BiSeNet (for face parsing), Dlib (for facial landmarks), Graph Attention Networks (GATs)

Author countries

China

← Previous