Recurrent-U-Net

Recurrent U-Net for Resource-Constrained Segmentation

Code

Please check the following URL for the code: https://github.com/kcyu2014/recurrent-unet/blob/master/README.md

Dataset

• KBH.tar.gz contains more than 12.5K annoated hand segmentation images.
• KBH2.tar.gz is an extension which contains 3,054 annoated hand segmentation images.
• In total, there are 15,590 annotated hand segmentation images.

Dataset Download Link

• KBH.tar.gz https://drive.google.com/file/d/1qEhfj7ezqzfzfJ_TbHC8PhJwzpg60bkN/view?usp=sharing
• KBH2.tar.gz https://drive.google.com/file/d/1oc4N78LQqSkY0d3x4E5Q-qqD0z_hLZpI/view?usp=sharing
• The two datasets need to be fused manually If you used this dataset in your research, please consider citing:

@inproceedings{wang2019recurrent,
  title={Recurrent U-Net for resource-constrained segmentation},
  author={Wang, wei and Yu, Kaicheng and Hugonot, Joachim and Fua, Pascal and Salzmann, Mathieu},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  pages={2142--2151},
  year={2019}
}

• Train Val Test Split

  • train_val_test_split folder saves the split used for the ICCV publication.
  • split.py shows how to get the train_val_test_split folder for KBH dataset.
  • split2.py shows how to get the train_val_test_split_ext folder for the extra images in KBH2.

• Command

  tar -xvzf KBH.tar.gz
  tar -xvzf KBH2.tar.gz
  python split.py
  python split2.py
  

  • txt fiels in train_val_test_split & train_val_test_split_ext folders need to be merged manually if you want to use all the images.