Course syllabus

The course aims to provide students with advanced knowledge and practical skills in computer vision, especially learning-based (deep learning) methods. Students learn how visual information is formed, represented and processed, and how to design, implement and evaluate computer vision pipelines in real-world applications relevant to machine learning, sensors and systems.

• Image formation and visual sensing, such as camera models, color spaces, noise, sampling, etc.
• Classical vision and representation, such as feature detection and description, geometric alignment, etc.
• Deep learning foundations for vision, such as CNNs, optimization, regularization, transfer learning, etc.
• Representation learning and architectures such as self-supervised and contrastive learning, vision transformers, etc.
• Core Vision Tasks such as image classification, object detection and tracking, semantic and instance segmentation, etc.
• Evaluation, robustness, and Ethics, such as metrics and benchmarks, model robustness and domain shift, bias, fairness, privacy in vision systems, etc.

On completion of the course, the student will be able to:

• explain key computer-vision problem formulations and the assumptions behind classical and learning-based methods.
• account for the main design choices in deep-learning-based vision systems.
• explain commonly used evaluation metrics and experimental protocols for major vision tasks.

On completion of the course, the student will be able to:

• implement and validate fundamental vision algorithms in a reproducible software workflow.
• train, debug, and evaluate deep neural networks for vision tasks using contemporary tooling, and apply transfer learning when appropriate.
• design an end-to-end vision pipeline for a given application, including dataset handling, model selection, evaluation, and reporting of results and limitations.
• communicate technical choices and results clearly in written form and orally, including quantitative comparisons and ablation-style reasoning.

On completion of the course, the student will be able to:

• critically assess scientific literature and implementations in computer vision with respect to assumptions, validity, robustness, and reproducibility.
• select methods that are appropriate for application constraints and justify the choices.
• reflect on ethical, legal, and societal implications of vision systems, including privacy risks and potential bias.