Video compression via deep learning for resource-constrained edge AI systems
Project Description
Video sensors have become ubiquitous and are playing critical roles in a variety of important applications, e.g., surveillance, video conferencing, vision-assisted robots, etc. Many of the devices are constrained in onboard computing resources and communication capability, which forms a major bottleneck for efficient video analytics tasks. This project will develop effective video compression methods via deep learning, and the compressed data will then be sent to be processed at powerful edge computing platforms.
Supervisor
ZHANG Jun
Quota
2
Course type
UROP1100
Applicant's Roles
The student will learn about deep learning based video compression and analytics methods, implement state-of-the-art methods on an edge AI platform, develop computationally efficient compression methods, and test the performance of various edge AI tasks. The applicant should have some background on machine learning.
Applicant's Learning Objectives
1. Study deep learning based video compression;
2. Implement start-of-the-art video compression methods;
3. Develop computationally efficient video compression methods;
4. Test the application of the proposed compression methods in edge AI applications.
Complexity of the project
Moderate