Wi-sc/BriNet

Few-shot Segmentation

Python

BriNet: Towards Bridging the Intra-class andInter-class Gaps in One-Shot Segmentation

By Xianghui Yang, Bairun Wang, Kaige Chen, Xinchi Zhou, Shuai Yi, Wanli Ouyang, Luping Zhou

Paper

You can find our paper at https://arxiv.org/abs/2008.06226

Citation

If you find BriNet useful in your research, please consider to cite:

@misc{yang2020brinet,
    title={BriNet: Towards Bridging the Intra-class and Inter-class Gaps in One-Shot Segmentation},
    author={Xianghui Yang and Bairun Wang and Kaige Chen and Xinchi Zhou and Shuai Yi and Wanli Ouyang and Luping Zhou},
    year={2020},
    eprint={2008.06226},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}

Training

You will also need to

Download/Prepare SBD dataset (http://home.bharathh.info/pubs/codes/SBD/download.html).
Download pre-trained ResNet50 from pytorch model zoo.

cd BriNet

# if you want to use default setting
python train.py -fold=0

# if you want to use default setting
python train.py -fold=0 -input_size=[353,353] -gpu=0 -checkpoint_dir='./checkpoint'

Testing

Download trained models (https://www.dropbox.com/sh/mt1mzr7sxq29he2/AADq83Y2IAGcO1swVo2fdslza?dl=0) or load your trained model. We assume you have downloaded the repository into ./checkpoint path.

cd BriNet

# if you want to use default setting
python test.py -fold=0

# if you want to use default setting
python test.py -fold=0 -input_size=[353,353] -gpu=0 -checkpoint_dir='./checkpoint'

Contact

For further questions, you can leave them as issues in the repository, or contact the authors directly: xianghui.yang@sydney.edu.au