SpaceTime Vision
Towards Unsupervised Learning in the 4D World
We live in a fast-developing and ever-changing world. Breakthrough advancements in deep learning, computer vision, and computational power enable today’s science to provide intelligent answers to the needs of our society. The capabilities of current artificial intelligence systems make people imagine and create smart applications in virtually all corners of human activity, impacting transportation, communication, entertainment, energy consumption, and the natural environment. While many automated solutions seem possible, there are still plenty of challenging questions remaining. Our main objective is to find practical answers to these scientific challenges.
Spacetime Vision Project pushes the frontiers of AI and takes important steps towards making the technologies of tomorrow possible, with direct impact in various parts of industry, energy, and the environment sectors. Our objectives are the following:
1) Create fast methods for online and unsupervised learning in large spatiotemporal volumes of data, able to function in the dynamic world. We will use different kinds of imaging and 4D (3D + time) sensing capabilities, ranging from fixed sensors to cameras present on UAVs. We will develop efficient methods for learning with no human supervision. Given the huge amounts of unlabeled data available and the costly manual annotation, unsupervised learning is crucial for the creation of new AI technologies.
2) Develop powerful methods for complete scene understanding, from the level of objects and activities involving objects to translating the visual scene into natural language. Vision to language translation is a new topic in AI, becoming a new exciting field of research.
3) Endow drones with the capacity to “see” and understand the world in which they fly. They will have the possibility to fly and land safely.
4) Develop smart cameras with new sensing and learning capabilities. They will make our research suitable for real-world applications.