Status: available Supervisor: Daniel Helm, Martin Kampel
Start: as soon as possible
Automatic scene analysis is a crucial task for historians or film archivists in order to preserve and interpret the memories of human cultural history. Film archives include thousands of hours of digitized analog footage from the last 100 years up to now and the manual preservation is a cost intensive and exhausting process for each individual expert. Additionally, archives include material with a broad diversity concerning quality, age of recording time, cinematographic techniques or film formats or different representations of similar content. Developing efficient scene analysis methods are mandatory in order to provide film experts innovative ways to search for specific scene content or identify abstract scene relations within large collections over time ranges up to 10 decades.
The goal of this practical work (or master thesis) is as follows:
The focus of this investigation is exploring and finding relations between images or image-sequences stored in a large film database. Therefore, fully unsupervised, supervised as well as semi-supervised learning strategies shall be explored in order to predict the corresponding similarity scores. Furthermore, different kinds of features such as global images features, object- or face-related features shall be explored.
- Literature Review – getting to know the algorithms (papers, github repos, …)
- State-of-the-art evaluation
- Creation of a usable dataset
- Implementation of own solution
- Evaluation (qualitative vs. quantitative)
- Readable and documented source-code
- Usable software package
- Final report + presentation
- Basic knowledge in computer vision
- Experience in Python.
- Interest in Deep Learning (Tensorflow, Keras, PyTorch) and Machine Learning
This work is part of the project “Visual History of the Holocaust”.
The practical course is part of an ongoing research project. A “Forschungsbeihilfe” is available for the selected student.