Photo: Huixia Liu
In his doctoral dissertation, Dingding Cai investigated how computers can determine an object's precise 3D position and orientation (6D pose) from a single colour or depth-enhanced image. The research introduced four deep-learning methods that better handle visually ambiguous objects, adapt from simulated training data to real-world scenes without costly pose labels, and estimate the poses of objects not used in training. The methods achieved leading results on benchmark datasets, including a system that works with only colour reference images and does not require pre-existing 3D CAD models, depth data or network retraining. The results can help developers and users of robots, augmented-reality applications and autonomous vehicles build systems that understand and interact with their surroundings more reliably.
The doctoral dissertation of M.Sc. (Tech.) Dingding Cai, in the field of Computing and Signal Processing, titled Single-View 6D Object Pose Estimation with Deep Learning, will be publicly examined at the Faculty of Information Technology and Communication Sciences at Tampere University on 28 August 2026.
The Opponent will be Assistant Professor Viktor Larsson from Lund University in Sweden. The Custos will be Professor Esa Rahtu from the Faculty of Information Technology and Communication Sciences, Tampere University.