×

Predictive Visual Completion of Simple Sketched Figures

Career / Placement Opportunities / BSc/MSc Topics / Predictive Visual Completion of Simple Sketched Figures
17 January. 2021
Deep neural-based architectures have emerged in the last decade as a powerful general-purpose substrate for learning directly from raw data, and have been used, in particular, for the predictive completion of images and text. It remains an interesting prospect whether shallower architectures, which presumably require considerably less training data, might be sufficient for the predictive completion of simple sketched figures. The project aims to develop an application for a touch-based device that allows a user to hand-draw simple figures, while the application attempts in parallel to anticipate and visualize the remainder of the figure. Depending on the internship type, the project can be adapted towards investigating theoretically whether simple local learning mechanisms and relatively few training examples suffice for the particular task, or towards the actual development of the application and the empirical evaluation of the efficacy of heuristic learning techniques.