MFFI: Multi-Dimensional Face Forgery Image Dataset for Real-World Scenarios

View on arXiv ← Back to list

Authors: Changtao Miao, Yi Zhang, Man Luo, Weiwei Feng, Kaiyuan Zheng, Qi Chu, Tao Gong, Jianshu Li, Yunfeng Diao, Wei Zhou, Joey Tianyi Zhou, Xiaoshuai Hao

Published: 2025-09-06 04:36:41+00:00

AI Summary

This paper introduces MFFI, a Multi-dimensional Face Forgery Image dataset designed to address the limitations of existing deepfake datasets in real-world scenarios. MFFI enhances realism across four strategic dimensions, integrating 50 different forgery methods and containing 1024K image samples. Benchmark evaluations demonstrate MFFI's superiority in scene complexity, cross-domain generalization, and detection difficulty.

Abstract

Rapid advances in Artificial Intelligence Generated Content (AIGC) have enabled increasingly sophisticated face forgeries, posing a significant threat to social security. However, current Deepfake detection methods are limited by constraints in existing datasets, which lack the diversity necessary in real-world scenarios. Specifically, these data sets fall short in four key areas: unknown of advanced forgery techniques, variability of facial scenes, richness of real data, and degradation of real-world propagation. To address these challenges, we propose the Multi-dimensional Face Forgery Image (\\textbf{MFFI}) dataset, tailored for real-world scenarios. MFFI enhances realism based on four strategic dimensions: 1) Wider Forgery Methods; 2) Varied Facial Scenes; 3) Diversified Authentic Data; 4) Multi-level Degradation Operations. MFFI integrates $50$ different forgery methods and contains $1024K$ image samples. Benchmark evaluations show that MFFI outperforms existing public datasets in terms of scene complexity, cross-domain generalization capability, and detection difficulty gradients. These results validate the technical advance and practical utility of MFFI in simulating real-world conditions. The dataset and additional details are publicly available at {https://github.com/inclusionConf/MFFI}.

Key findings

MFFI significantly outperforms existing public datasets in terms of scene complexity, cross-domain generalization capability, and detection difficulty gradients. Spatial-domain detectors (e.g., Xception, RFM) demonstrated superior robustness to real-world degradations on MFFI compared to frequency-domain counterparts. Even advanced multi-modal large models (MLLMs) exhibited limited performance on MFFI, highlighting the significant real-world challenges retained by the dataset.

Approach

The paper addresses limitations in existing deepfake detection datasets by proposing MFFI, a new multi-dimensional face forgery image dataset. MFFI is constructed to enhance realism across four strategic dimensions: wider forgery methods (encompassing 50 types), varied facial scenes, diversified authentic data from multiple sources, and multi-level degradation operations, including adversarial attacks, to simulate real-world propagation.

Datasets

MFFI (proposed); Authentic data sources for MFFI: CelebA, RFW (Racial Faces in the Wild), CASIA-WebFace, Chinese Celeb; Datasets used for comparison/benchmarking: DF-TIMIT, FaceForensics++ (FF++), Celeb-DF, DeeperForensics-1.0, DFDC, ForgeryNet, FakeAVCeleb, DF3, DeepFakeFace, DiffusionDeepfake, DF40, CDF-V1, CDF-V2, DFD.

Model(s)

UNKNOWN

Author countries

China, UK, Singapore

← Previous