CAGNet

This repository contains the Tensorflow implementation of our paper "CAGNet: Content-Aware Guidance for Salient Object Detection". The paper can be found at: [Sciencedirect] [arXiv]

Requirements

• numpy 1.17.4
• tensorflow 2.1.0
• scipy 1.4.1
• pillow 7.0.0
• opencv-python

Usage

1- Clone the repository

git clone https://github.com/Mehrdad-Noori/CAGNet
cd CAGNet

2- If you want to train the model, download the following dataset and unzip it into data folder.

• DUTS Dataset: Training

3- To run the training, set the arguments or use the default settings:

python train.py --backbone_model 'ResNet50' --batch_size 10 --save_dir 'save'

The backbone_model can be one of the following options: VGG16, ResNet50, NASNetMobile or NASNetLarge

4- To generate saliency maps:

python predict.py --model '/path/to/trained/model' --input_dir /path/to/input/images/directory --save_dir 'save'

You can also download and use our pre-trained models

5- Evaluation code

You can use this toolbox to compute different saliency measures such as E-measure, Weighted F-measure, MAE, PR curve ...

Pre-trained models & pre-computed saliency maps

We provide the pre-trained model and pre-computed saliency maps for DUTS-TE, ECSSD, DUT-OMRON, PASCAL-S, and HKU-IS datasets.

• CAGNet-V (VGG16 backbone) - model: GoogleDrive, baidu (extraction code: bne8)- saliency maps: GoogleDrive, baidu (extraction code: m3id)
• CAGNet-R (ResNet50 backbone) - model: GoogleDrive, baidu (extraction code: bp22) - saliency maps: GoogleDrive, baidu (extraction code: kv9w)
• CAGNet-M (NASNet Mobile backbone) - model: GoogleDrive, baidu (extraction code: 8kx6) - saliency maps: GoogleDrive, baidu (extraction code: q8v1)
• CAGNet-L (NASNet Large backbone) - model: GoogleDrive, baidu (extraction code: idh8) - saliency maps: GoogleDrive, baidu (extraction code: kety)

Quantitative Comparison

Qualitative Comparison

Any problems?

Please feel free to contact me, or raise an issue if you encounter any problems.

Citation

@article{mohammadi2020cagnet,
  title={CAGNet: Content-Aware Guidance for Salient Object Detection},
  author={Mohammadi, Sina and Noori, Mehrdad and Bahri, Ali and Majelan, Sina Ghofrani and Havaei, Mohammad},
  journal={Pattern Recognition},
  pages={107303},
  year={2020},
  publisher={Elsevier}
}