Modern Deep Learning has fundamentally changed Artificial Intelligence. Long-held dreams (and concerns) are now gradually becoming reality with Deep Learning being applied almost everywhere, esp. to the problems of image and video understanding, retrieval, and synthesis. Several recent research demos have demonstrated to even outperform human performance in difficult applications such as generation of high-resolution natural images or classification of cancerous tissue in medical imagery.
The algorithms and the fundamental research in deep Machine Learning and Computer Vision that are driving this revolution are improving at an ever-increasing rate. The goal of this seminar is, therefore, to provide students with an overview over the latest research in this area. Besides taking an in-depth look at state-of-the-art publications, we will also facilitate the discussion of their shortcomings. Moreover, we will provide an opportunity to brainstorm about potential remedies for these limitations, which can serve as the basis for a future practical or a final thesis in a later semester.
Topics include but are not limited to:
* Image & video synthesis
* Visual superresolution and Image completion
* Artistic style transfer
* Interpretability of deep models
* Visual object classification, detection, tracking
* Self-supervised learning
* Metric and representation learning
* Modern deep learning approaches, such as transformers and self-attention, invertible neural networks, etc.
Process
We will provide a list of papers and each student will select one. Students have to read and understand the paper as well as related background information, present the ideas of the paper in the seminar, and provide a report. In addition, all students should engage in the discussions of the papers presented by others in the seminar.
- المعلم: Ursula Fantauzzo
- المعلم: Timo Milbich
- المعلم: Björn Ommer
- المعلم: Matthias Wright