GC-ConsFlow: Leveraging Optical Flow Residuals and Global Context for Robust Deepfake Detection

Authors: Jiaxin Chen, Miao Hu, Dengyong Zhang, Jingyang Meng

Published: 2025-01-23 07:43:56+00:00

AI Summary

This paper proposes GC-ConsFlow, a novel dual-stream framework for robust Deepfake video detection that integrates spatial and temporal features. It addresses challenges posed by realistic manipulated videos and interference from natural facial motions by capturing complementary spatiotemporal forgery traces. The method outperforms state-of-the-art techniques, demonstrating effectiveness and robustness, especially under various compression scenarios.

Abstract

The rapid development of Deepfake technology has enabled the generation of highly realistic manipulated videos, posing severe social and ethical challenges. Existing Deepfake detection methods primarily focused on either spatial or temporal inconsistencies, often neglecting the interplay between the two or suffering from interference caused by natural facial motions. To address these challenges, we propose the global context consistency flow (GC-ConsFlow), a novel dual-stream framework that effectively integrates spatial and temporal features for robust Deepfake detection. The global grouped context aggregation module (GGCA), integrated into the global context-aware frame flow stream (GCAF), enhances spatial feature extraction by aggregating grouped global context information, enabling the detection of subtle, spatial artifacts within frames. The flow-gradient temporal consistency stream (FGTC), rather than directly modeling the residuals, it is used to improve the robustness of temporal feature extraction against the inconsistency introduced by unnatural facial motion using optical flow residuals and gradient-based features. By combining these two streams, GC-ConsFlow demonstrates the effectiveness and robustness in capturing complementary spatiotemporal forgery traces. Extensive experiments show that GC-ConsFlow outperforms existing state-of-the-art methods in detecting Deepfake videos under various compression scenarios.

Key findings

GC-ConsFlow achieved superior performance on FF++ under HQ compression across all forgery types and competitive results under LQ compression, outperforming most existing methods. The model also demonstrated strong generalization ability on the Celeb-DF dataset, surpassing several state-of-the-art baselines. Ablation studies confirmed the effectiveness of the GGCA module and the use of optical flow residuals for enhanced spatial and temporal modeling.

Approach

The GC-ConsFlow framework uses a dual-stream approach: a global context-aware frame flow (GCAF) stream with a Global Grouped Context Aggregation (GGCA) module for spatial artifact detection, and a flow-gradient temporal consistency (FGTC) stream. The FGTC stream leverages optical flow residuals and gradient-based features to robustly detect temporal inconsistencies while mitigating interference from natural facial movements.

Datasets

FaceForensics++ (FF++), Celeb-DF

Model(s)

XceptionNet (backbone for GCAF and FGTC), FlowNet (for optical flow computation)

Author countries

China

← Previous