In this theme we look at new techniques that can bring two images into dense correspondence. This includes stereo matching, optical flow and scene flow computation. We currently focus on the following research challenges. Firstly, we consider large displacement motion, such as in sport events. Secondly, we would like to answer the question whether additionally information, such as the recovered material and lighting, helps to perform image matching better. Thirdly, going a step further we would like to take two noisy and blurry images and derive a generative model of the image formation process. We call this task inverse rendering.
- A. Sellent, C. Rother, S. Roth. “Stereo Video Deblurring”, ECCV 2016. [project][preprint]
- H. Abu Alhaija, A. Sellent, D. Kondermann, C. Rother. “GraphFlow – 6D Large Displacement Scene Flow via Graph Matching”, German Conference on Pattern Recognition (GCPR, a.k.a. DAGM), 2015. [project][pdf]
- M. Hornacek, F. Besse, J. Kautz, A. Fitzgibbon and C. Rother, Highly Overparameterized Optical Flow Using PatchMatch Belief Propagation, ECCV 2014.
- L. Torresani, V. Kolmogorov, and C. Rother, A Dual Decomposition Approach to Feature Correspondence, PAMI 2013.
- F. Besse, C. Rother, A. Fitzgibbon, and J. Kautz, PMBP: PatchMatch Belief Propagation for Correspondence Field Estimation, BMVC 2012 – Industrial Impact Prize award.
- M. Bleyer, C. Rother, P. Kohli, D. Scharstein, and S. Sinha, Object Stereo – Joint Stereo Matching and Object Segmentation, CVPR 2011.