Implementation of Bilateral Space Video Segmentation [Maerki et al CVPR 2016]
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


By Oliver Wang

This is a personal reimplementation of the method proposed in Bilateral Space Video Segmentation [Maerki et al. 2016]. Importantly, it was written for clarity of understanding, and is totally unoptimized (i.e. IT'S SLOW and uses a LOT of memory). Furthermore, it only partially implements the published method, featuring only multi-linear interpolation. For realistic timing and results to compare against, please refer to the material included in the original publication.

All code provide here is to be used for research purposes only. For questions regarding commercial use, including licensing rights and IP, you must contact the owner; The Walt Disney Company.

Note: If you use this software, please cite the following work:

Bilateral Space Video Segmentation
Nicolas Maerki, Federico Perazzi, Oliver Wang, Alexander Sorkine-Hornung
2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)


This method also uses a third party GraphCut library. Please refer to the webpage for the most up to date information.


The datasets included originate from the FBMS-59 dataset. The datasets are provided only for research purposes and without any warranty. When using the BMS-26 or FBMS-59 in your research work, you should cite the appropriate papers in the link above.

C++ Implementation

Scott Wehrwein implemented a version of BVS in C++ for his project on foreground/background segmentation. You can find it here.