Neural networks are increasingly used in many practical applications, including image recognition and classification. However, there are two important limitations to the process of neural network training. Firstly – the training requires a very large set of image data. Secondly – the objects in these images must be manually tagged. The process requires a human to recognize objects and mark their contours in order for the neural network to be able to learn what a given photo depicts and to generalize this knowledge. The automation of this process is limited in this way.
Sky Engine offers a solution to those limitations. The concept of the Sky Engine platform is based on the observation that modern 3D graphics rendering systems are able to generate extremely realistic images that can potentially be used to train neural networks in order to recognize or classify objects. Using a single 3D model, it is possible to generate a virtually unlimited number of images depicting objects of interest – from different points of view, on different backgrounds, under different lighting conditions, and in varying ranges of the light spectrum – it is possible, for example, to simulate X-ray images, infrared light or multispectral images. The combination of automatic generation of image data and deep learning modules into one platform enables unprecedentedly fast training of neural networks. Thanks to this approach, time and costs can be minimized while ensuring higher accuracy of inference. In addition, Sky Engine offers automation of the creation of the 3D models.
Sky Engine technology has a number of practical applications. One of them is the interpretation of images from medical diagnostic systems. In the field of endoscopy, for example, the Sky Engine platform enables the rapid training of neural networks to recognize cancer and other diseases. Sky Engine solutions also prove themselves very useful in the analysis of visual data in sport. During the Covid19 pandemic, they also support the interpretation of data from city cameras in order to monitor social distancing. In addition, the Sky Engine technology can be used for automatic inspection of: electronic components, parts and components in production processes, solar panels, farmland, metal parts or medical equipment.
Sky Engine solutions have already found their first customers. Scania AB (Sweden) uses the Sky Engine analytics system to optimize its truck service revenues. Land Robots (UK) uses Sky Engine technology to train drones and robots for work in vineyards. Smart Tracking (PL) uses Sky Engine solutions in the system for supporting decision-making in volleyball refereeing.
The key members of the Sky Engine team are:
– Jakub Pietrzak (CTO) – an experienced specialist in computer vision systems, machine learning algorithms development, data processing and graphics processors;
– Ph.D. phys. eng. Bartek Włodarczyk (CEO) – clinical medical physicist, specialist in the field of oncological radiotherapy and diagnostic medical imaging, with over sixteen years of experience in the field of medical physics in clinical and technical applications; also has business experience with the leaders of the global radiotherapy market.
The key partners of Sky Engine are: Amazon Web Services, Nvidia, Microsoft, IBM, Adobe, Allegorithmic, Oregon State University (US), University of Warsaw, Land Robots, Golem, BritLyt, TechData, NIAB EMR (UK).