This repository contains the reference code for our TMM paper: arXiv Paper Version

If you use any part of our code, or DINet is useful for your research, please consider citing::

@article{yang2019dilated,
  title={A dilated inception network for visual saliency prediction},
  author={Yang, Sheng and Lin, Guosheng and Jiang, Qiuping and Lin, Weisi},
  journal={IEEE Transactions on Multimedia},
  volume={22},
  number={8},
  pages={2163--2176},
  year={2019},
  publisher={IEEE}
}

• Python 2.7
• Keras 2.1.2
• Tensorflow-gpu 1.3.0
• opencv-python

• Clone this repo:

git clone https://github.com/ysyscool/DINet
cd DINet
mkdir models

• Download weights from Google Drive. Put the weights into

cd models

Download the SALICON 2015 dataset and modify the paths in config.yaml And then using the following command to train the model

python main.py --phase=train --batch_size=10

For testing, modify the variables of weightfile (in line 217) and imgs_test_path (in line 220) in the main.py. And then using the following command to test the model

python main.py --phase=test

Please refer to this link.

Code largely benefits from sam.