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

http://www.zhaoliming.net/research/deepsaliency

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

Contact: Liming Zhao ([zhaoliming@zju.edu.cn](mailto: zhaoliming@zju.edu.cn))

Paper

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.

Dependencies

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

Usage

• 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.

Training

• Download dataset to dataset directory
• Run dataset\create_caffe_data.py to obtain the hdf5 training data
• Then training using the script in models\finetune.sh