FANet

The repository is for the paper "FANet: Features Adaptation Network for 360$^{\circ}$ Omnidirectional Salient Object Detection", IEEE Signal Processing Letters, 2020.

Codes

• The code is trained and tested with Python3.7, PyTorch1.6 and CUDA10.1. The required packages include PyTorch, torchvision, Numpy, SciPy, PIL, OpenCV and Tensorboard.

• The pretrained weight of backbone ResNet-50 can be downloaded from official PyTorch link. The datasets can be downloaded from 360-SOD and F-360iSOD.

• The paths in the config.yaml should be reset when you need to train the model or predict the saliency maps.

• The eval code can be found in http://dpfan.net/.

Results

Our results can be downloaded at results or BaiduYun CloudDrive(Extraction Code: alab).

Citation

If you find this repo useful, please cite the following paper:

@ARTICLE{Huang_2020_SPL,
  author={M. {Huang} and Z. {Liu} and G. {Li} and X. {Zhou} and O. {Le Meur}},
  journal={IEEE Signal Processing Letters}, 
  title={FANet: Features Adaptation Network for 360$^{\circ}$ Omnidirectional Salient Object Detection}, 
  year={2020},
  volume={27},
  pages={1819-1823},
  doi={10.1109/LSP.2020.3028192}}

Contact

Any questions, please contact huangmengke@shu.edu.cn.