Box-supervised Instance Segmentation via Level Set Evolution

Wentong Li, Wenyu Liu, Jianke Zhu, Miaomiao Cui, Xiansheng Hua and Lei Zhang
Paper (arXiv). ECCV2022.

News

• Inspired by our boxlevelset, BoxSnake is developed. Please see the paper for the details.

• We develop BoxInstSeg, which is a toolbox for box-supervised instance segmentation. More methods (e.g. BoxInst, DiscoBox, Box2Mask) are updated in it. This survey repo will update the methods on box-supervised instance segmentation.

• We extend our work to Box2Mask to obtain the powerful performance, which build our method on CNN-based (Box2Mask-C) and Transformer-based (Box2Mask-T) frameworks. Please see our paper (arXiv) for details. The codes and pretrained models for Box2Mask are released in BoxInstSeg.

Visual performance on COCO of our Transformer-based method(Box2Mask-T).

Model Zoo

• This code is for our eccv version paper. The pretrained models are in model_zoo.

• The codes and pretrained models for Box2Mask are availble in BoxInstSeg.

Installation

This implementation is based on MMdetection. Please refer to install.md for detailed installation.

Getting Started

Please see getting_started.md for models training and inference.

Citation

@inproceedings{li2022box,
  title={Box-supervised Instance Segmentation with Level Set Evolution},
  author={Wentong Li, Wenyu Liu, Jianke Zhu,  Miaomiao Cui,  Xiansheng Hua and Lei Zhang},
  booktitle={European Conference on Computer Vision},
  pages={1--18},
  year={2022}
}

@article{li2022box2mask,
  title={Box2Mask: Box-supervised Instance Segmentation via Level Set Evolution},
  author={Wentong Li, Wenyu Liu, Jianke Zhu,  Miaomiao Cui, Risheng Yu, Xiansheng Hua and Lei Zhang},
  journal={arXiv},
  year={2022}
}

More

This repo will update the survey of box-supervised instance segmentation.

License

This project is licensed under the Apache License 2.0.