Sevilla-Lara et al. largely build upon the work by Sun et al.  by incorporating semantic segmentation into the proposed layered model for optical flow. Furthermore, the approach by Sun et al. is slightly altered to what is called "localized layers". Their approach is based upon the following two requirements:
Given the semantic segmentation, the initial flow field is refined depending on the class:
At the time of publication, Sevilla-Lara et al. reported the leading result on the KITTI dataset , see the leaderboard, and qualitative results look promising on both the KITTI dataset and on selected Youtube sequences, see Figure 1.
What is your opinion on the summarized work? Or do you know related work that is of interest? Let me know your thoughts in the comments below or using the following platforms: