Here we see the Cortexica Deep Neural Network demonstration. This demo shows the reliability of using Neural Networks for image recognition, in this case accelerated using OpenCL running on the GPU. The use of GPU Compute allows the algorithm to run using the powerful Arm® Mali®-T628 GPU together with the Arm Cortex®-A15 & A7 CPUs. The deep neural network in this example, has been trained offline and is capable of recognising over 700 images, showing great resilience to image size, position, rotation and quality.
By using Arm technology, Cortexica have been able to move the processing from servers in the cloud to the mobile platform itself. This significantly reduces bandwidth and increases response time. The Cortexica algorithm was a straight port from Nvidia® CUDA® to OpenCL.
This demo runs on the Samsung Chromebook 2 powered by the Exynos 5422 processor, with the Cortex-A15 & A7 CPUs in a big.LITTLE™ configuration and a Mali-T628 GPU MP6.
You can find more information about GPU Compute on Mali here: http://malideveloper.arm.com/
For more information about Cortexica: www.cortexica.com