Design and implementation of an DL-based video pocessing pipeline

Abstract

[en] This project was born as an internal need at Amazon for being able to automatically or semi-automatically perform Quality Assurance checks to videos featuring books in order to eliminate or reduce the human time spent on doing video QA. The proposed workflow uses a variety of tools and algorithms that will be detailed, including the use of deep learning. This is a good example of a real-life use-case in which deep learning is used to help improve the humans life. This video processing pipeline is composed of some checks to ensure that the videos are correctly rendered and is already integrated to the major architecture Waldo Media Factory, and ready to be used in production.