Lecture Course: Deep Learning for Graphics and Visualization, Winter Term 2020/21

Illustration: serveral Visualisations based on Deep Learning applications

Objectives

Machine Learning can be found in almost all fields of computer science. This course teaches basic concepts of machine learning and how they are applied to computer graphics. This course covers the whole process of developping, training neural nets and also adapting complex models to new datasets. Learning from 3D points aka. point clouds as it is covered in this course, is a current research topic in the field of computer graphics. Students will thus first learn how to solve standard machine learning problems, before applying their know how to 3D data. All practical realizations will be made in Tensorflow, which is also introduced in the course.

We assume previous knowledge in computer science, but not necessarily in machine learning.

Content

  • Basics: Regression vs. Classifications, Supervised vs. Unsupervised, Gradient Descent
  • Multi-Layer Perceptrons
  • Training Concepts: Back Propagation, Optimizers, Overfitting, Regularization
  • Introduction to TensorFlow 2
  • Convolutional Neural Networks: 2D / 3D Convolution, Pooling
  • Advanced Architectures: ResNet, DenseNet, Attention Mechanisms, Recurrent Models, Encoder-Decoder, Autoencoder
  • Generative Models
  • Unstructured CNNs: Multi-View, PointNet, Poisson Pooling
  • Visualization for Neural Networks

Exercises

The exercise lessons will be held interactively through BBB.

Lecture

The lecture will take place digitally through Moodle and BBB.

Possible Allocation

Computer Sciences

  • B.Sc. Main Subject
  • M.Sc. PCS

Media Informatics

  • B.Sc. Main Subject
  • M.Sc. Media Informatics

Software Engineering

  • M.Sc. PCS

Cognitive Systems

  • Sp. Sub.: Perception

(725 00/136 82)