Main page
Lund University

Open PhD position

Deep Learning for 3D Computer Vision

The topic of this project is to develop new learning methods and mathematical models for the next generation of autonomous systems capable of understanding, navigating and mapping in 3D environments. In recent years, the performance of 2D recognition has dramatically improved due to Deep Learning, but today’s best-performing systems for 3D computer vision are based on purely geometric concepts, not using any recognition. There is no single, integrated modelling framework leveraging both geometry and semantics. The research will be focused on mathematical tools appropriate for the above challenge, hence a strong mathematical background is required. The work is relevant for many industrial applications, such as self-driving cars, augmented reality and autonomous robots.

We are looking for highly talented and creative PhD candidates to join our research team. There will be several researchers (PhD students, a post-doc and three professors) actively involved in this research direction of autonomous, perceptual systems, ranging from strong mathematicians to more applied, industrial partners. As an example, we are closely collaborating with Zenuity AB on self-driving cars.

Application deadline: 31 May 2017. Apply here and follow the instructions. Note that you must write in your application that you are interested in Project C2: Deep Learning for 3D Computer Vision.

Paper: See our latest work on deep structured models and Markov Random Fields: Learning Arbitrary Pairwise Potentials in CRFs for Semantic Segmentation where we achieve state-of-the-art results on the challenging CityScapes benchmark. Joint work with Phil Torr's group at University of Oxford.

Paper: See our latest work on 3D reconstruction using Lagrangian duality: Why Rotation Averaging is Easy. Joint work with Queensland University of Technology.

See also publications.

Publisher: Fredrik Kahl,
Last updated: May 4, 2017