Skip to content

'Bi-directional Relationship Inferring Network for Referring Image Segmentation' CVPR2020

Notifications You must be signed in to change notification settings

fengguang94/BRINet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

57 Commits
 
 
 
 
 
 
 
 

Repository files navigation

BRINet for referring image localization and segmentation

This repository contains code for:

'Bi-directional Relationship Inferring Network for Referring Image Segmentation', CVPR 2020.

'Bidirectional Relationship Inferring Network for Referring Image Localization and Segmentation', TNNLS 2021

If you find this work useful in your research, please consider citing:

@inproceedings{hu2020bi,
  title={Bi-directional relationship inferring network for referring image segmentation},
  author={Hu, Zhiwei and Feng, Guang and Sun, Jiayu and Zhang, Lihe and Lu, Huchuan},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={4424--4433},
  year={2020}
}
@article{feng2021bidirectional,
  title={Bidirectional Relationship Inferring Network for Referring Image Localization and Segmentation},
  author={Feng, Guang and Hu, Zhiwei and Zhang, Lihe and Sun, Jiayu and Lu, Huchuan},
  journal={IEEE Transactions on Neural Networks and Learning Systems},
  year={2021},
  publisher={IEEE}
}

Paper link

The paper can be found in paper.

Code

Requirement

Setup

Partial coda and data preparation are borrowed from TF-phrasecut-public. Please follow their instructions to make your setup ready. DeepLab backbone network is based on TF-deeplab as well as the pretrained model for initializing weights of our model.

Sample code

Training

python main_cmsa.py -m train -w deeplab -d Gref -t train -g 0 -i 800000

Testing

python main_cmsa.py -m test -w deeplab -d Gref -t val -g 0 -i 800000

About

'Bi-directional Relationship Inferring Network for Referring Image Segmentation' CVPR2020

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published