Professional: ICON
Industry Cooperation
The central objective of the project is the development of methods in the area of data-driven processing of multimedia content. This is achieved through the exploration, application, and adaptation of methods from the domain of artificial intelligence and machine learning with the aim of utilizing them for data-driven applications in a multimedia environment. Many core algorithms in the realm of multimodal image and video processing are currently in a nascent stage of development. Therefore, a preliminary industrial research phase is essential for their deployment in an economic context. However, the methods to be developed potentially support various use cases, all of which are based on automated identification, metadata extraction, and algorithmic exploitation of big data multimedia content. This project was a collaboration with Pro7Sat1 Media SE and MMI Munich Media Intelligence GmbH.
In this project, I was responsible for the machine learning models concerning object detection and action recognition. I evaluated pertinent open-source models using our internal datasets and contributed to the development of pipelines. Furthermore, this project led me to delve into face recognition. We also participated in the Person Recognition challenge (WIDER person challenge @ ECCV2018 in Munich), securing 4th place. The challenge centered around identifying actors in every frame of a movie, including views from behind and other challenging angles. What did I learn? I became acquainted with the agile methodology of SCRUM. I also had the opportunity to work with JIRA. The close collaboration offered me a deep insight into the company's processes and workflows.
Technologies