Scale-Space for Machine Learning on 3d point clouds – SSLAM

Some decades ago, the advent of digital photography led to the formation and development of dedicated research fields (e.g., image processing, pattern recognition, computer vision), which have revolutionized our daily lives. We believe that, in a near future, 3d acquisition will have a similar impact than digital photography. The ambition of this project is to develop and demonstrate a framework for the interactive analysis and understanding of acquired 3d point clouds with billions of points, a task that is out of reach of existing techniques. We will take advantage of the strong expertise I developed in the past years on adapting the Scale-Space analysis on 3d point clouds. Recently, we proposed to classify edge-points by combining Scale-Space computations with a dedicated neural network. This new idea leads to a huge improvement of the classification quality, speed, frugality and energy consumption.
In this project, we will build upon this proposition and explore the association of Scale-Space and ML for extracting geometric and semantic knowledge from complex acquired point cloud of billions of points. Our goal is to speed up the Scale-Space computation and define more advanced ML models. We will focus on fast and tractable approaches, and study the design of systems interactively driven by the user expertise and the application specificity (e.g., interactive learning). We will study how «to advance the state of the art in order to achieve complex tasks (e.g., pattern recognition, computer vision), and to allow high-level interactions with Human users», a theme mentioned in the call of the AAPG, axe E.2. Our project focuses on «learning from unstructured data» for applications related to «Computer Vision» and «Pattern Recognition». We want to extend the Scale-Space model for «knowledge extraction and representation» from unstructured data. Beside, we will study interactive learning for 3d point clouds, considering Networks architectures, HCI, and HPC aspects.