Source Code and Model for DeepSaliency: Multi-Task Deep Neural Network Model for Salient Object Detection
Branch: master
Clone or download

DeepSaliency Project

DeepSaliency is a saliency object detection framework based on fully convolutional neural networks with global input (whole raw images) and global output (whole saliency maps).

###Project Website

In the project website, you can find detailed descriptions, models, result maps and datasets used in our paper.

Contact: Liming Zhao ([](mailto:


Xi Li, Liming Zhao, Lina Wei , Ming-Hsuan Yang, Fei Wu, Yueting Zhuang, Haibin Ling, Jingdong Wang. "DeepSaliency: Multi-Task Deep Neural Network Model for Salient Object Detection“. IEEE Transactions on Image Processing (TIP), 2016.


  • Caffe (included in the project code)
  • Python (ipython notebook is used)
  • Linux (Windows is also OK with modification)


  • Download or clone the project code
  • In the models directory, download the models from google drive
  • Then a demo for processing one input image can be found in demo.ipynb.


  • Download dataset to dataset directory
  • Run dataset\ to obtain the hdf5 training data
  • Then training using the script in models\