FCANet

The official PyTorch implementation of CVPR 2020 paper "Interactive Image Segmentation with First Click Attention".

Requirement

PyTorch>=0.4.1
Opencv
Scipy
Matplotlib

Useage

Put the pretrained model into the folder "pretrained_model" and the unzipped datasets into the folder "dataset".

Evalution:

python evaluate.py --backbone [resnet,res2net] --dataset [GrabCut,Berkeley,DAVIS,VOC2012] (--sis)

Demo with UI:

python annotator.py --backbone [resnet,res2net] --input test.jpg --output test_mask.jpg (--sis)

([x,y,z] means choices. (x) means optional.)

Datasets

GrabCut ( GoogleDrive | BaiduYun pwd: 6qkj )
Berkeley ( GoogleDrive | BaiduYun pwd: 6bfw )
DAVIS ( GoogleDrive | BaiduYun pwd: u5vd )
(The three datasets we used are downloaded from f-BRS)
PASCAL ( official )

Pretrained models

fcanet-resnet ( GoogleDrive | BaiduYun pwd: g989 )
fcanet-res2net ( GoogleDrive | BaiduYun pwd: xzmx )

Other framework

We offer the code in constructed by MindSpore framework.

Citation

If you find this work or code is helpful in your research, please cite:

@inproceedings{lin2020fclick,
  title={Interactive image segmentation with first click attention},
  author={Lin, Zheng and Zhang, Zhao and Chen, Lin-Zhuo and Cheng, Ming-Ming and Lu, Shao-Ping},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={13339--13348},
  year={2020}
}

Contact

If you have any questions, feel free to contact me via: frazer.linzheng(at)gmail.com.
Welcome to visit the project page or my home page.

License

The source code is free for research and education use only. Any comercial use should get formal permission first.

frazerlin / fcanet