Deep Learning for Computer Vision

This advanced course covers the principles, techniques, and applications of deep learning in computer vision. Students will learn how to develop, train, and validate neural networks for image classification, object recognition, semantic segmentation, and other computer vision tasks. Techniques for improving the performance of deep learning models and illustrations will also be covered to provide clues for further model development. By the end of the course, students will be able to apply deep learning techniques to solve real-world problems in various domains.

The course is managed via the Moodle learning platform. All documents and further information can be found in the Moodle course Deep Learning for Computer Vision SoSe2024.

Please notice: The materials for our lectures and exercises are only available through the network of the Bauhaus-Universität Weimar.

Integrated lecture

Lectures

  1. Organisation, history and perceptron
  2. Optimization and regularization
  3. Convolutional neural networks
  4. Image classification and transfer learning
  5. Architectures
  6. Transformer
  7. Object detection
  8. Semantic and instance segmentation
  9. Probabilistic generative models
  10. Deep learning for image matching
  11. 3D deep learning applications I
  12. 3D deep learning applications II

Assignments

  1. Backpropagation and Python project management
  2. Data loading, model architecture, training and evaluation
  3. Kaggle, real training times, network fitting and empirical work
  4. Modular network design, tensor shapes and object detection

Project

  1. Phase 1. individual project part
  2. Phase 2. team project part
  3. Project presentation

Exam

Written exam

  • Date: July 29, 2024, 9:00 Uhr
  • Place: SR 3.31 at S143
  • Auxiliary resources: none

Preparation material

  • Old exam samples