Organisational information

Please register by e-mail to david.schneider@kit.edu, stating your first name, last name, matriculation number and the examination regulations (Prüfungsordnung, SPO) under which you are studying.

Content

Current human-machine interfaces are still largely "blind" when it comes to perceiving their users. They are therefore neither able to use natural human communication channels such as facial expressions, gaze direction, gestures or body language for human-machine interaction, nor to gain sufficient knowledge about their users, their state and their intentions. Current research aims to improve this and to develop new human-machine interfaces that perceive their users and their actions, and that use the context information gained in this way to interact with the users appropriately.

In this seminar, the participants study and present recent work from the following areas:

• Face detection and recognition
• Facial expression recognition
• Estimation of head pose, gaze direction and attention
• Localization and tracking of people
• Person identification
• Articulated body tracking and body model estimation
• Gesture recognition
• Interactive segmentation of medical images
• Medical image segmentation (anatomy and pathology in CT, label quality, evaluation metrics)
• Visual in-context learning and learning with few annotations (semi-, weakly- and self-supervised learning)
• Multi-modal medical agents
• Document analysis (layout and structure analysis, robustness, chart understanding)
• Visual localization and scene segmentation (panoramic, open-vocabulary)
• 3D scene understanding and vision-language models for assistive technology
• Fine-grained action understanding (temporal action segmentation, human-object interaction)
• Computer vision for robotics

Each participant gives a seminar talk in English (25-30 min) on the chosen topic, followed by a discussion, and writes a report. The report of about 5-10 pages is only due at the end of the semester, but we recommend preparing it before the talk if possible. Each participant is expected to become familiar with the topic independently and to research further literature.

For Master's students, the assessment is an examination of another type. The final grade is composed in equal parts of the grades for the report, the presentation and the participation in the seminar (1/3 each).

Assessment

The assessment consists of a written report and its presentation, as an examination of another type according to § 4 (2) No. 3 SPO.

Prerequisites and recommendations

• Knowledge of the fundamentals of pattern recognition, as taught in the core module Cognitive Systems [IN3INKS / IN4INKS], is required.
• The seminar complements the lecture of the same name. We therefore recommend attending the lecture as well.

Language of instruction

English