- Poster sessions are a terrific spot to learn and demonstrate.
- One example: How to quickly process LiDAR data into a useful immersive research tool.
- Reaching beyond STEM to humanities scholars a new norm among HPC applications.
The student poster sessions at tech conferences are always an exciting place to visit. There, the future is written in big letters; the possibilities that innovation brings to society charge the air.
One of these forward looking posters at the XSEDE16 conference described the creation of a High Performance Visualization Pipeline (HPVP) to process data-intensive LiDAR point cloud. Developed by a team at Virginia Polytechnic Institute and State University (Virginia Tech), HPVP promises assistance to researchers in both STEM and humanities domains.
LiDAR (Laser Imaging, Detection and Ranging) is a surveying method that employs lasers to measure distantly lit targets. It is a technique widely used in mapping, with airborne and terrestrial applications for atmospheric physics, archaeology, seismology, forestry, and more – basically any place you’d want a detailed map of an area you’d like to study.
LiDAR is especially helpful when mapping areas not accessible to GPS-fed devices, areas such as dense forest, interiors, or agricultural areas.
Trouble is, LiDAR generates a lot of data in the process, so turning it into a visual, explorable form is computationally costly.
“Several hundred megabytes can be feasibly visualized using standard tools on a single powerful machine,” says Ayat Mohammed. “However, LiDAR scans of even small environments can easily lead to massive datasets consisting of billions of points. Rendering these massive datasets with standard visualization tools often exceed the machine's main and graphics memory.”
To resolve this bottleneck, Mohammed’s team from Virginia Tech devised HPVP to render, visualize and interactively navigate these massive datasets. Mohammed’s team was able to quickly preprocess a five-billion-point cloud of LiDAR data in quick order using Virginia Tech’s BlueRidge supercomputer. BlueRidge has a total of 6,528 cores and 27.3 TB of memory.
After preprocessing, the team converted the point cloud files into a format readable by the ParaView visualization tool, an open source application designed specifically for data and visualization analysis. After that the team partitioned the files using ParaView's D3 filter.
The next step was to render the ParaView files remotely, and finally they converted the data into an immersive 3D environment. An immersive 3D representation encourages a more intuitive and interactive exploration of data. Researchers interested in underground caverns, cumulonimbus banks, archeological digs, and more will benefit from an immersive 3D experience that until now was difficult to obtain.
Offering a lead into supercomputing for humanities scholars is one of the main motivations for developing the tool, Mohammed stresses.
"To use HPVP, researchers need to scan the place where they want to conduct the experiment. Then they can get a manageable 3D model of their experiment space that can be displayed as an immersive virtual environment," Mohammed says. "This is helpful for studies that have a possible risk such as fear of heights, for instance. We had a similar experiment carried out by cognitive science graduate students for a class project. The researchers wanted to test for word memorization in a virtual environment that mimics the real experiment environment."
Mohammed's visualization tool is one of those insightful supercomputing applications to be seen at tech showcases. Her HPVP reaches beyond the traditional STEM domains to offer humanities scholars an accessible entry point into HPC-powered research. It's because of researchers like Mohammed and her team at Virginia Tech — and because of organizations like XSEDE who foster these budding computer scientists — that computationally-enabled discoveries are accelerating.