This was a course project for NCSU's Computer Science 791-025: Real-Time AI & High-Performance Machine Learning. The goal of this project was to accelerate a video super resolution model for mobile devices. State of the art compression and optimization techniques were used to compress our model, accelerate it, and optimize it for mobile devices.
Running Super Resolution for video is a difficult task to do in real-time, and it's even more challenging for a resource constrained device such as a smartphone. You not only have to consider the compute and memory constraints of a mobile device, but also the battery life.
Slides presented are available to view in slides.pdf. The final report is available in final.pdf.
We targeted two smartphone devices for benchmarking and development. The first is the Samsung Galaxy S10e and the other is the iPhone 13 Pro Max. For training, we used the NVIDIA V100 GPU nodes at Bridges-2 supercomputer.
To simply run everything, just run this command:
python3 final.pyIf you want to run on bridges-2, then just submit a batch job:
sbatch job-gpuThere are a lot of comamnds associated with this project to learn how to use them do
python3 final.py -hThe original source code is available to view here.