Authors: Yaning Zhang, Zitong Yu, Xiaobin Huang, Linlin Shen, Jianfeng Ren

Published on: February 03, 2024

Impact Score: 8.27

Arxiv code: Arxiv:2402.02003

Summary

• What is new: Introduces a large-scale, diverse, and high-fidelity dataset, GenFace, for deepfake detection, including advanced diffusion-based generated faces. Proposes a novel detector, CAEL, enhanced with an AECA module for better forgery identification.
• Why this is important: The challenge of distinguishing between real and manipulated images due to advancements in photorealistic generators, particularly with the emergence of diffusion model-based forgeries which are underrepresented in current datasets.
• What the research proposes: A new dataset, GenFace, with a wide range of advanced forged images for benchmarking, along with a new detection model employing cross appearance-edge learning and an appearance-edge cross-attention module for improved forgery detection.
• Results: The proposed detection model demonstrates superior performance in detecting deepfakes across various evaluation settings, including different generators and manipulation techniques.

Technical Details

Technological frameworks used: nan

Models used: Diffusion-based models, Cross Appearance-Edge Learning (CAEL) detector, Appearance-Edge Cross-Attention (AECA) module

Data used: GenFace dataset

Potential Impact

Social media platforms, digital forensics services, cybersecurity firms, content creation industries

Want to implement this idea in a business?

We have generated a startup concept here: AuthentixVision.