Description

In the last years, many efforts have been made to retrieve the 3D reconstruction of an object captured by a single camera. While a wide variety of solutions have exploited motion cues to solve that, the use of pixel intensities has been constrained to laboratory setups. However, it is worth noting that this information is a key factor to recover non-specific detail-aware complex objects. We aim to propose a new method to capture 3D objects along with illumination properties from pictures. To that end, we could consider two alternatives: 1) a neural-based solution with surface modelling and 2) a fully-interpretable variational approach with physical priors. What do you prefer? One of them could be your final thesis project!

Requirements:
# Background: Computer Science, Mathematics, Telecommunications and Industrial Engineering
# Other: Experience in computer vision and machine learning algorithms will be valued
# Language: Fluent in English