Research Interests: James Gain

A/Prof. Gain is looking for Masters and Doctoral students in the following areas but is willing to consider students with their own topics in Computer Graphics, Visualization, and High Performance Computing:

Software Tools for 3D Computer Animation

We currently collaborate with Triggerfish Animation Studios and this involves a number of applied research projects (and associated student funding):

1. Distributed Simulation for Visual Effects - Simulation of natural phenomena such as water, fire and object collision and fracture is widespread in the Visual Effects industry. However, such simulation tends to be extremely compute intensive especially as the number of simulated particles is increased in order to ensure realism. Visual effects studios are often under intense time pressure and these simulations would thus benefit from execution on a cluster of computers. Fortunately, such clusters are already in use as “render farms”. This project seeks to parallelise simulations in order to create an effective “simulation farm”.
2. Agent-based Crowds - currently agents forming a crowd (such as in the Massive software utilised for battle scenes in Lord of the Rings) can be steered in certain directions using flow fields or by increasing the probabilities of certain behaviour. We would like to give directors more specific control, to the extent of specifying an agent's position or state at a certain time. In general such constraint specification is a difficult problem as it involves specifying some of the terminal states in a large search space. Of course, this can benefit from heuristic methods of state space search developed in Artificial Intelligence. We have undertaken some promising preliminary work in this area but there is much still to do from the perspective of ensuring that the resulting tools are suited to animator control.
3. Procedural Methods - Procedural methods are an effective way of generating complex and realistic geometric scenes with little user input. They constitute a set of algorithmic techniques (fractals, L-systems, shape grammars and noise functions) that iteratively build structures like trees, terrain, road networks and buildings and are widely used for Games, Visual Effects and Virtual Environments. Unfortunately, the very fact that they require little user input works against these methods since they also afford little user control. One promising route for future research is to incorporate better control through sketch-based and example-driven interfaces. Another avenue is to apply procedural methods in new areas, such as mechanical assemblies.
4. Retargeting Assets for Computer Games - A substantial proportion of revenue from computer animated films comes from tie-in products such as computer games. Here it is important to be able to re-purpose existing digital assets, in particular character models, for the much more constrained environments typical of interactive and mobile games. Although simplifying geometric models and creating different levels of detail [12] has been a staple of computer graphics for many years, the focus here would be on allowing artists to easily specify attributes of particular visual importance on the model and have these respected during the simplification. The broad research question is thus: how can film-quality assets be automatically converted into those suitable for real-time mobile games?
   

Heritage Data

The Geomatics department at UCT is involved in an initiative to create an integrated database of architectural cultural heritage sites in Africa. The condition of these sites is deteriorating and given their importance in understanding and interpreting humankind's history, a permanent digital record is essential.

It is common practice in Geomatics to digitally capture the shape of buildings for analysis purposes, using laser range scanning. This produces a cloud of points that represent the position and colouration of samples on the building surface.

There are a number of projects arising from this rich data-set:

1. Automatic Hole-Filling - Unfortunately, due to constraints on the position of the laser scanner and the convolutions of the building point clouds often have gaps. An automated approach to filling these holes is to transplant complete surfaces from elsewhere in the model, using the edges around the hole as context.
2. Certainty Visualisation - Archaeologists are particularly concerned with any error that might arise during the capture process. Given the diversity of possible data sources (with scans from different devices at different times), it is worth portraying these uncertainties visually. The challenge here is to avoid overly distorting the normal rendering of the building and its textures.
3. Texture Matching - Scanning often produces two sets of data: digital photographs (texture) and laser scans (point clouds). For a truly effective representation of geometry and texture these data sets need to be registered but this can be challenging, particularly if the exact position and orientation of the scanning device is not known.
4. Display in Virtual Environments - In a digital museum setting it is worth attempting to convey something of the experience of seeing the original heritage site. Many of these sites, such as churches and fortresses, have a significant vertical component, which virtual environments traditionally represent poorly.

Computational Astronomy

The High Performance Simulation and Visual Computing Laboratory has a multi-year collaborative project to investigate scalable solutions to deal with the processing of radio spectrum data as produced by the South African MeerKAT radio telescope (and in future the SKA). Projects in this area can be targeted towards either Visualization or Graphical Processing Unit (GPU) acceleration of existing astronomy algorithms, such as pulsar detection.