Machine learning explores the study and construction of algorithms that can learn from and make predictions on data. In computer vision this technology enables the cognition of the enviroment through an optical sensor. In this way devices can see and understand their surroundings like humans. Now, devices can recognize objects, identify people and understand an overall scene and its context.
Among the most promising Machine Learning technologies is the Deep Learning technology which is based on artificial neural networks composed of many layers. Based on the power of multilayer CNNs, this category of machine learning revolutionizes computer vision applications, by offering high recognition rates and robustness needed to many critical applications.
In order to optimally port state-of-the-art computer vision algorithms to a specific platform and specializing on the development of on-device software for embedded systems, IRIDA Labs relies heavily on heterogeneous computing techniques. As such are characterized a set of programming methods which optimally exploit any available on-device computing resource, in order to achieve optimal algorithm implementation through optimal system partitioning.
By exploiting these techniques, IRIDA Labs has successfully carried out a variety of porting projects for various high-end computing platforms.