Bay Vision - Spring 2018

Mohamed Musa is the founder and CEO of Deepen AI . The company provides AI based models, tools, and services for developing autonomous systems. “ The main intention of autonomous products is to bring safety, reliability and productivity to the world -- Musa says -- we want to save human lives and give back people time ; instead of being stuck in traffic, autonomy will solve most of these issues and allow mobility to others that couldn’t previously move around very often, such as the disabled and blind. The impact that autonomy can bring extends past just mobility and parking lots: if you don’t need parking spaces, the whole design of the city and infrastructure can change significantly and that can give us more space. It’s a huge impact to the world, better for the environment and better for humanity. We want to realize that dream ! ” Deepen AI is currently at growth stage with ten full time staff and 50+ labeling resources, while additionally recruiting new employees. The range of products they work with includes: cars, drones, robots including anything that has a vision system and tries to make intelligent decisions about mobilizing in the world. One product has previously been released and already multiple customers are utilizing their products and services. The future plans for the company involve expanding the product offering and to hopefully collaborate with everybody in the autonomous product field. Their expertise is based on 3D computer vision, such as LiDAR, sensor fusion, radar and camera together, stereo cameras and different types of the vision sensors. Since every product out there should at least include a camera, Deepen AI aims to work together at some capacity with any company that has a camera on their system, from car manufacturers, tier-1 suppliers, sensor producers to robot companies. The computer vision techniques that have been most helpful up to this stage include a heavy reliability on deep learning: they use a lot of neural networks of all different types, from RCNNs to regulars CNNs and LSTMs . They don’t try to include deep learning for every issue, unless it is a good fit for the problem. Simultaneously, they recognize that certain aspects are better done with optical flow and will try to utilize anything that solves the problem at hand. “ Within the deep learning techniques, semantic segmentation has been challenging for the team at Deepen AI -- Musa declares -- you can always get OK classification with bounding boxes but when it gets to pixel level accuracy, there is still a lot of work to reach that level of granularity.” The industry currently relies on manual labor “The main intention of autonomous products is to bring safety, reliability and productivity to the world.” 8 Boston Vision Deepen Ai