Machine Learning Based Video Compression
|Project title||Machine Learning Based Video Compression|
|Background||At its core, a video compressor's purpose is to find a binary representation of an array of pixels that uses less bits than the buffer that contains those pixel values. Contemporary implementations use a variety of algorithms to create the compressed binary representation.
Video encoding is a billion dollar industry and current algorithms are currently encumbered by a number of patents. A revolutionary approach to video compression, such as this, could have major commercial implications.
|Description||The goal of this project will be to use machine learning to create an algorithm that can encode to and decode from a binary representation of an array of pixels and compare this to well-known algorithms such as JPEG.
The project would consist of components that could be split between team members, including application design, video processing and ML algorithm development.
|Deliverable||An encoding application that will take an uncompressed video file and create a compressed video file, and a decoding application that will take a compressed video file and create an uncompressed video file.|
|Skill set desirable||Python / TensorFlow for the ML algorithm, and C++ / C# for the application logic. Telestream have a team of three software developers who will be available to advise. No previous knowledge of video compression is assumed.|
|Client time availability||30-60 min weekly or more|
|IP requirement||Open source project|