Unsupervised Online Video Object Segmentation with Motion Property Understanding

Tao Zhuo, Zhiyong Cheng*, Peng Zhang*, Yongkang Wong, and Mohan Kankanhalli

Results on DAVIS-2016 TrainVal Dataset (50 videos)

Measure	NLC	LMP	FSEG	ARP	UOVOS
J Mean	64.1	69.7	71.6	76.3	77.8
J Recall	73.1	82.9	87.7	89.2	93.6
J Decay	8.6	5.6	1.7	3.6	2.1
F Mean	59.3	66.3	65.8	71.1	72.0
F Recall	65.8	78.3	79.0	82.8	87.7
F Decay	8.6	6.7	4.3	7.3	3.8
T	36.6	68.8	29.5	35.9	33.0

NLC: Video Segmentation by Non-Local Consensus voting. A. Faktor, M. Irani, BMVC 2014.
LMP: Learning Motion Patterns in Videos. P. Tokmakov, K. Alahari, C. Schmid, CVPR 2017.
FSEG: FusionSeg: Learning to combine motion and appearance for fully automatic segmentation of generic objects in videos. S. Jain, B. Xiong, K. Grauman, CVPR 2017.
ARP: Primary Object Segmentation in Videos Based on Region Augmentation and Reduction. Y.J. Koh, C.-S. Kim, CVPR 2017.

Setup

Ubuntu
Matlab
Python2.7
Opencv_3.4
Mask-RCNN https://github.com/matterport/Mask_RCNN

Citation

If you use this code, please cite the following paper:

@article{zhuo2018unsupervised, title={Unsupervised Online Video Object Segmentation with Motion Property Understanding}, author={Zhuo, Tao and Cheng, Zhiyong and Zhang, Peng and Wong, Yongkang and Kankanhalli, Mohan}, journal={arXiv preprint arXiv:1810.03783}, year={2018} }

Contact

Tao Zhuo (zhuotao@nus.edu.sg)