Blazingly Fast Video Object Segmentation with Pixel-Wise Metric Learning

This is a demo of our work 'Blazingly Fast Video Object Segmentation with Pixel-Wise Metric Learning', presented at CVPR 2018. The work aims to learn a pixel-wise embedding, where similar pixels are close in the embedding space, for video object segmentation. More details can be found in the paper.

If you find it helpful for your research, please consider citing the paper:

@inproceedings{chen2018blazingly,
  title={Blazingly Fast Video Object Segmentation with Pixel-Wise Metric Learning},
  author={Chen, Yuhua and Pont-Tuset, Jordi and Montes, Alberto and Van Gool, Luc},
  booktitle = {Computer Vision and Pattern Recognition (CVPR)},
  year={2018}
}

If you encounter any problem, feel free to contact me at yuhua[dot]chen[at]vision[dot]ee[dot]ethz[dot]ch.

Installation

Clone the repo.

git clone https://github.com/yuhuayc/fast-vos.git

The implementation is based on the Caffe version from PSPNet. Clone the repo, and build Caffe and matcaffe (see: Caffe installation instructions)
```
git clone https://github.com/hszhao/PSPNet.git
make -j8 && make matcaffe
```
Create a soft link of Caffe in the project folder.
```
cd $VOS_ROOT/
ln -s $CAFFE_ROOT caffe
```

Interactive Video Segmentation Demo

An example script is provided in 'demo_interactive.m'.
Pre-trained model can be downloaded at here, move it to $VOS_ROOT/models/interactive/