Pruning in face forgery detection
With the rise of deep learning and modern technology, Deepfake algorithms can be used to easily forge fake faces that are so realistic and difficult to distinguish by the human eye. This poses a significant threat to information security as attackers have more power to control and manipulate one’s i...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2023
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/165852 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | With the rise of deep learning and modern technology, Deepfake algorithms can be used to easily forge fake faces that are so realistic and difficult to distinguish by the human eye. This poses a significant threat to information security as attackers have more power to control and manipulate one’s identity to achieve their malicious purposes. To combat this increasingly relevant issue, multiple state-of-the-art Face Forgery detection algorithms have been developed, and many have achieved impressive results when put to the test. However, these algorithms often utilise deep Convolutional Neural Networks (CNN) and the high complexity and computational resources required makes it difficult for one to implement them in a production environment.
In this project, we researched on multiple network pruning methodologies and applied these findings on a publicly available Face Forgery detection algorithm known as LipForensics. With the aim of compressing the model’s size while maintaining its accuracy and performance, we hope to achieve speedup of in model inference time and decrease in the model size. |
|---|