Syllabus

  • Students will get to know the current methods and technologies in the field of three-dimensional computer graphics.
  • Emphasis will be given to rendering different types of data: volumetric data, point clouds, mesh geometry, and logically defined geometry in the fields of medicine, biology, geodesy, and high-energy physics.
  • Because the rendered data can be very large, emphasis will also be given to the application of appropriate algorithms and data structures for fast and real-time rendering, implementation of techniques on graphic processors, and remote rendering.
  • The students will get to know the benefits of modern graphics libraries (Vulkan, WebGPU) for addressing these challenges.
  • In addition to the techniques, the students will also get acquainted with the different ways of visualizing such data, how to utilize different deep learning tools on the data for visualization preparation or visual parameter estimation and how to select suitable visualization method for an individual domain.
  • Students will have an opportunity to collaborate and interact with students and staff from one of the world’s best Visual computing groups in the world at KAUST.

Assignments

Visualization of related entities in historical records: More and more collections of historical records are becoming publicly available, opening up an interesting area of historical investigation. In this context, we are interested in how we can automatically extract different entities (persons, places, events, etc.) from such corpora, connect them to each other and display this in an intuitive way. The goal of such a task is the automatic generation of a wide variety of static or interactive, preferably online, visualizations that meaningfully display the found relations and entities. Visualization of Geodetic data This research topic focuses on the semantic segmentation of large-scale geodetic point clouds to identify and classify urban features such as buildings, vegetation, and roads. The task involves evaluating and applying recent deep learning methods tailored for point cloud data to enable accurate object classification, counting, and spatial distribution analysis. Emphasis is placed on adapting these models to the characteristics of geodetic datasets, which often include varying densities and noise levels. The results will be visualized using lightweight techniques to support qualitative evaluation and comparison across urban areas. Possible applications road and building conditions, or providing inputs for digital twins and sustainable urban planning strategies. Visualization of Astronomy data This research topic explores the use of deep learning techniques for detecting and classifying celestial objects in large-scale multispectral datasets obtained from ESA and NASA space telescopes. The focus is on processing high-dimensional data from instruments such as Hubble, JWST, or Gaia to identify stars, galaxies, nebulae, and potential transient phenomena across different wavelengths. The task includes adapting convolutional and transformer-based models to handle the noise characteristics, varying resolutions, and occlusions present in astronomical images. Visualization will support qualitative inspection and facilitate the identification of uncertain or rare detections. Potential applications include automated object cataloguing, assisting with the discovery of new astronomical phenomena, supporting time-domain astronomy, and contributing to large-scale sky surveys and mission planning. Visualization of cellular structures Recently, more and more collections of segmented microscopic data of subcellular structures have appeared, which are segmented using different (mostly deep) approaches. The goal of the task is to develop an online tool for comparing the performance of different approaches, where the user can upload segmented structures with an individual approach, and then review the performance of their segmentation and their mutual comparison in 2D and 3D views. Such a tool should also contain as many analytical and comparative tests as possible and enable the production of direct visualizations on individual cases, as well as the visualization of the results of such analyzes in the form of graphs and tables. Statistical modeling of the uterus shape The shape and size of human organs can vary considerably between individuals. The same applies to the case of uteruses, where variations in size are small, but there are differences in their shape and position in the body. The goal of the task is to develop a statistical model of the shape of the uterus, based on already segmented uteruses, which can be used to determine and visualize deviations from the average shape and size of the uterus. In addition to the developed statistical model, the goal is also to develop a tool for visualizing such a comparison. Hybrid server-user visualization of high energy physics data In high energy physics (e.g. LHC experiments at CERN), a huge amount of collision data is typically generated for each experiment. The aim of the task is to develop a hybrid server-user online visualization system for displaying such data, where the visualization of a large amount of less critical data is handled by the server, and the most important data is drawn directly by the user. With this, we want to enable advantage of the computing power of server systems.

  • There are numerous other challenges available.