The official PyTorch implementation of oral paper "FocusCut: Diving into a Focus View in Interactive Segmentation" in CVPR2022.

See requirements.txt for the environment.

pip3 install -r requirements.txt

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

CUDA_VISIBLE_DEVICES=0 python main.py -rf -ap "backbone='resnet50'"

CUDA_VISIBLE_DEVICES=0 python main.py -v -r focuscut-resnet50.pth -ap "backbone='resnet50'" -hrv -dv GrabCut,Berkeley,DAVIS 

CUDA_VISIBLE_DEVICES=0 python annotator.py -r focuscut-resnet50.pth -ap "backbone='resnet50'" -hrv  -img test.jpg

(These datasets are organized into a unified format, our Interactive Segmentation Format (ISF)

• GrabCut ( GoogleDrive | BaiduYun )
• Berkeley (GoogleDrive | BaiduYun )
• DAVIS (GoogleDrive | BaiduYun )
• SBD (GoogleDrive | BaiduYun )

• focuscut-resnet50 (GoogleDrive | BaiduYun )
• focuscut-resnet101 (GoogleDrive | BaiduYun )

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

@inproceedings{lin2022focuscut,
  title={FocusCut: Diving into a Focus View in Interactive Segmentation},
  author={Lin, Zheng and Duan, Zheng-Peng and Zhang, Zhao and Guo, Chun-Le and Cheng, Ming-Ming},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={2637--2646},
  year={2022}
}

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.

Licensed under a Creative Commons Attribution-NonCommercial 4.0 International for Non-commercial use only. Any commercial use should get formal permission first.