Build a video app like Hotstar/Netflix/Tiktok at low cost using deep learning
One of the major problems faced by a company doing video-on-demand or live streaming live Hotstar, Netflix etc. is the video distribution costs via CDN and is one of the major pain points in scaling the company.
To deal with this I will be demonstrating a solution of reducing the bandwidth costs of streaming using AI. We will be looking at a technique called Super-Resolution which can upscale image resolution without making it look like its blurry.
Then I will demonstrate how super-resolution can be applied for videos by making use of redundancy across frames in a video and how it can be dynamically applied when the frames get decoded from a compression algorithm say H264 or VP9.
Outline/Structure of the Demonstration
- Problem Statement i.e Video Streaming costs and existing solutions(2 min)
- Solution and Intro to Super-Resolution(2 min)
- Brief of various techniques which can used for super-resolution(5 min)
- Super-resolution for video(3 min)
- Content aware super-resolution(2 min)
- Metrics to measure video quality and the effectiveness of super-resolution(2 min)
- How to deal with client-side issues(2 min)
- Q & A(2 min)
After attending this session the attendees will
Get an understanding of how video streaming works
Get an understanding of Super-resolution and it's implementations
Get an understanding of Super-resolution for video and how it reduces CDN costs
Computer Vision Engineers
Prerequisites for Attendees
What is deep learning and how neural network works
What is a CNN
CNN training concepts