2021 Fiscal Year Annual Research Report
Facial Privacy and Forensic in The Wild: Explainable End-to-End Networks for Multi-Face Anonymization and Multi-Face Forgery Detection
| Project/Area Number |
20K23355
|
|---|
| Research Institution | National Institute of Informatics |
|---|
Principal Investigator |
レ チュンギア 国立情報学研究所, 情報社会相関研究系, 特任研究員 (00884404)
|
|---|
| Project Period (FY) |
2020-09-11 – 2022-03-31
|
| Keywords | deepfake generation / deepfake detection / deepfake segmentation |
|---|
| Outline of Annual Research Achievements |
We developed a forgery workflow to reduce the cost of synthesizing fake data. Our framework can generate an infinite number of fake individual
identities using GAN models for non-target face-swapping without repeatedly training a deepfake AE. This framework has great potential in
deepfake generation and face anonymization. We also created a new large-scale dataset with high-quality images for multi-face forgery detection and segmentation in-the-wild. It consists of 115K unrestricted images with 334K human faces. We also presented a benchmark suite to facilitate the evaluation and advancement of these tasks. Our work was published at ICCV 2021, a top-tier conference in computer vision.
We wrote a book chapter to introduce deepfake for beginners. Our chapter overviewed deepfake generation and detection methods from the viewpoint of technical evolution in computer vision. In particular, it described deepfake generation methods in different categories and analyzed their limitations. In addition, it clarified the different tasks of deepfake detection, from conventional classification to modern end-to-end detection and segmentation. It further discussed the limitations of deepfake detection methods and suggested solutions for improving the robustness of deepfake detection. Finally, it suggested a future direction for deepfake detection. This book chapter is in publishing process.
|
Research Products
(9 results)
