Semantic 3D segmentation is a task essential to applications that require an understanding of real-world 3D scenes, such as robotics, artificial intelligence (AI), augmented or virtual reality (AR/VR), and autonomous navigation/driving. The successful candidate is expected to conduct research at the intersection of computer vision, computer graphics and machine learning, by integrating concepts and methods from these areas to advance the state of the art in 3D scene understanding.
Basic knowledge of computer vision and deep learning. Programming skills: python, TensorFlow (optional), PyTorch (optional)
The scientific objectives of the project span a range of topics from these research areas, including data collection, neural networks training, evaluation, and application development, with the final goal to develop a novel deep learning architecture for semantic 3D segmentation, composed of deep neural networks for segmenting and labelling real-world objects and scenes.
Expected Deliverables: Final report, trained networks, code base