Anstarc / HDNet

code for paper "HDNet: Hybrid Distance Network for semantic segmentation"

HDNet: Hybrid Distance Network for semantic segmentation

A deep neural network with attention mechanism for semantic segmentation task.

 

 

• Segmentation results on PASCAL Context

  

 

• Application on skin detection task

  

 

Requirements

• Install PASCAL in Detail
• Install requirements.txt

How to use

• To train HDNet:

python train.py --dataset pcontext_detail --out_dir /out_dir --pretrained_home /pretrained_home --data-folder /data-folder 

The model and log are saved in --out_dir

• To test HDNet:

python test.py --dataset pcontext_detail --resume-dir /resume-dir --data-folder /data-folder --pretrained_home /pretrained_home --eval --multi-scales

 

Citation

if you find HDNet useful in your research, please consider citing:

@article{li2021hdnet,
  title={HDNet: Hybrid Distance Network for semantic segmentation},
  author={Li, Chunpeng and Kang, Xuejing and Zhu, Lei and Ye, Lizhu and Feng, Panhe and Ming, Anlong},
  journal={Neurocomputing},
  volume={447},
  pages={129-144},
  year={2021},
  publisher={Elsevier}
}

Acknowledgement

Thanks for DANet, PyTorch-Encoding, timm