In this theme we develop efficient inference techniques for 3D Scene understanding. Our ultimate goal is to take a few RGB(D) images and output in real-time the full scene-graph, with all objects present in the scene, their corresponding attributes and their 3D spatial relationship, e.g. “object A is supported by object B”. While this is very hot research area with many ongoing efforts, we currently focus on few research directions. These are in particular: 3D Pose estimation of known object instances or classes and semantic segmentation of (stereo) images.
- Object Instance Recognition and Pose Estimation (jointly with Prof Gumhold’s team (TUD))
- Dense Semantic Segmentation with Objects and Attributes (this project is run by Shuai Zheng from Phil Torr’s team in Oxford)
- 3D Semantic Segmentation and Tracking on an airfield (supporting Prof. Fricke’s team (TUD). DFG Project)
- E. Brachmann, F. Michel, A. Krull, M. Y. Yang, S. Gumhold, C. Rother, Uncertainty-Driven 6D Pose Estimation of Objects and Scenes from a Single RGB Image, CVPR 2016.
- F. Michel, A. Krull, E. Brachmann, M. Y. Yang, S. Gumhold, C. Rother, Pose Estimation of Kinematic Chain Instances via Object Coordinate Regression, BMVC 2015
- A. Krull, F. Michel, E. Brachmann, S. Gumhold, S. Ihrke, C. Rother, 6-DOF Model Based Tracking via Object Coordinate Regression, ACCV 2014 (the associated system won a demo honorable mention award at ACCV 14)
- E. Brachmann, A. Krull, F. Michel, S. Gumhold, J. Shotton, and C. Rother, Learning 6D Object Pose Estimation using 3D Object Coordinates, ECCV 2014
- V. Vineet, C. Rother, P. H.S. Torr, Higher Order Priors for Joint Intrinsic Image, Objects, and Attributes Estimation, NIPS 2013
- M. Bleyer, C. Rhemann, and C. Rother, Extracting 3D Scene-consistent Object Proposals and Depth from Stereo Images, ECCV 2012