Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DiffFace-Edit: A Diffusion-Based Facial Dataset for Forgery-Semantic Driven Deepfake Detection Analysis

Feng Ding, Wenhui Yi, Xinan He, Mengyao Xiao, Jianfeng Xu, Jianqiang Du

TL;DR

This work targets the gap in forgery datasets for fine-grained, region-level facial edits and the challenge posed by detector-evasive samples. It introduces DiffFace-Edit, a diffusion-based dataset with over 2 million partially edited faces across eight facial regions and three manipulation types, including detector-evasive samples and precise prompt annotations. A cross-domain IMDL benchmark with eight locators is used to study generalization under semantic ambiguity and model perturbations. Findings show detector-evasive and richly edited samples markedly reduce localization accuracy, underscoring the need for cross-domain evaluation to develop robust forensic tools that generalize beyond single-model assumptions.

Abstract

Generative models now produce imperceptible, fine-grained manipulated faces, posing significant privacy risks. However, existing AI-generated face datasets generally lack focus on samples with fine-grained regional manipulations. Furthermore, no researchers have yet studied the real impact of splice attacks, which occur between real and manipulated samples, on detectors. We refer to these as detector-evasive samples. Based on this, we introduce the DiffFace-Edit dataset, which has the following advantages: 1) It contains over two million AI-generated fake images. 2) It features edits across eight facial regions (e.g., eyes, nose) and includes a richer variety of editing combinations, such as single-region and multi-region edits. Additionally, we specifically analyze the impact of detector-evasive samples on detection models. We conduct a comprehensive analysis of the dataset and propose a cross-domain evaluation that combines IMDL methods. Dataset will be available at https://github.com/ywh1093/DiffFace-Edit.

DiffFace-Edit: A Diffusion-Based Facial Dataset for Forgery-Semantic Driven Deepfake Detection Analysis

TL;DR

This work targets the gap in forgery datasets for fine-grained, region-level facial edits and the challenge posed by detector-evasive samples. It introduces DiffFace-Edit, a diffusion-based dataset with over 2 million partially edited faces across eight facial regions and three manipulation types, including detector-evasive samples and precise prompt annotations. A cross-domain IMDL benchmark with eight locators is used to study generalization under semantic ambiguity and model perturbations. Findings show detector-evasive and richly edited samples markedly reduce localization accuracy, underscoring the need for cross-domain evaluation to develop robust forensic tools that generalize beyond single-model assumptions.

Abstract

Generative models now produce imperceptible, fine-grained manipulated faces, posing significant privacy risks. However, existing AI-generated face datasets generally lack focus on samples with fine-grained regional manipulations. Furthermore, no researchers have yet studied the real impact of splice attacks, which occur between real and manipulated samples, on detectors. We refer to these as detector-evasive samples. Based on this, we introduce the DiffFace-Edit dataset, which has the following advantages: 1) It contains over two million AI-generated fake images. 2) It features edits across eight facial regions (e.g., eyes, nose) and includes a richer variety of editing combinations, such as single-region and multi-region edits. Additionally, we specifically analyze the impact of detector-evasive samples on detection models. We conduct a comprehensive analysis of the dataset and propose a cross-domain evaluation that combines IMDL methods. Dataset will be available at https://github.com/ywh1093/DiffFace-Edit.
Paper Structure (10 sections, 2 figures, 6 tables)

This paper contains 10 sections, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. DiffFace-Edit dataset
  3. Pristine Data Collection and Partially Synthetic Face Generation
  4. Dataset Statistics
  5. Benchmark Settings
  6. Results and Analysis
  7. Evasion Analysis in IMDL
  8. Counterintuitive Analysis of Removal Samples
  9. Cross-Manipulation Experiments
  10. Conclusion

Figures (2)

  • Figure 1: The fake data generation pipeline of proposed DiffFace-Edit dataset.
  • Figure 2: Distribution of the splicing subset. Panel (a) shows the proportion of each edited region in single-region edits; panel (b) shows the proportion of each edited region in multi-region edits; panel (c) shows the distribution of multi-region edits by number of edited regions.