Better Artificial Vision using Nvidia GPUs and PS3 gaming devices
It seems like video game consoles and technology is being increasingly used in the field of medicine and research. And researchers from MIT and Harvard have demonstrated how to build better artificial visual systems using low-cost, high-performance gaming hardware.
The neural processing which goes into visually recognizing, even the simplest object in an environment is huge, and extremely difficult to recreate or mimic. A huge amount of the inner workings of biologically based systems remains a mystery. The researchers from MIT and Harvard have registered progress, faster than before, utilizing Graphics Processing Units (GPUs).
“We made a powerful computing system that delivers over hundred fold speed-ups relative to conventional methods,” explained Nicolas Pinto, a PhD candidate in James DiCarlo’s lab at the McGovern Institute for Brain Research at MIT. “With this extra computational power, we can discover new vision models that traditional methods miss.” Pinto has co-authored the PLoS study with David Cox of the Visual Neuroscience Group at Harvard’s Rowland Institute.
Here’s how the team did it. Using the processing power of dozens of high-performance Nvidia graphics cards and PS3 gaming devices, the team first designed a high-throughput screening process to obtain the best parameters for visual object recognition tasks. The model which resulted performed way better than a “crop” of state-of-the-art vision systems across a range of tests. Specifically, this model was able to identify a range of objects against random natural backgrounds with a variation in position, scale and rotation, with greater accuracy. Additionally, the MIT release suggests that, had the team used conventional methods, the one week screening phase would have ended up taking over two years to complete.
The researchers say that this high-throughput approach could also be applied to various other areas of computer vision, such as face recognition, object tracing, gesture and action recognition, and pedestrian detection for automotive applications. Their model will also help scientists to better understand the human brain.
