Integrated Variational And Nearest Neighbor field (IVANN) for Optical Flow

Zhuoyuan Chen, Ying Wu, Hailin Jin, Zhe Lin, Scott Cohen

Aug 31, 2016 (modified: Aug 31, 2016) ECCV2016 BNMW submission readers: everyone
  • Submit for proceedings: yes
  • Abstract: It is a fundamental problem to construct accurate dense correspondences between two images. Despite the efforts and promising methods handling relatively small motion, one remaining challenge is induced by large and complex non-rigid motion. Aiming at this challenge, the new method proposed exploits the mutual boosting between the variational flow and the nearest-neighbor field (NNF). The proposed method “IVANN” gives a very effective solution under rather complex motion, and currently achieved state-of-the-art performance on both the Middlebury[3] and MPI- Sintel benchmarks[7].
  • Conflicts: baidu.com, northwestern.edu, adobe.com, inria.fr, tum.de

Loading