SalFBNet

This repository includes Pytorch implementation for the following paper:

SalFBNet: Learning Pseudo-Saliency Distribution via Feedback Convolutional Networks, 2021. (pdf)

Guanqun Ding, Nevrez Imamoglu, Ali Caglayan, Masahiro Murakawa, Ryosuke Nakamura

Citation

Please cite the following papers if you use our data or codes in your research.

@misc{ding2021salfbnet,
      title={SalFBNet: Learning Pseudo-Saliency Distribution via Feedback Convolutional Networks}, 
      author={Guanqun Ding and Nevrez Imamouglu and Ali Caglayan and Masahiro Murakawa and Ryosuke Nakamura},
      year={2021},
      eprint={2112.03731},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

@inproceedings{ding2021fbnet,
  title={FBNet: FeedBack-Recursive CNN for Saliency Detection},
  author={Ding, Guanqun and {\.I}mamo{\u{g}}lu, Nevrez and Caglayan, Ali and Murakawa, Masahiro and Nakamura, Ryosuke},
  booktitle={2021 17th International Conference on Machine Vision and Applications (MVA)},
  pages={1--5},
  year={2021},
  organization={IEEE}
}

Getting Started

1. Installation

You can install the envs mannually by following commands:

conda create -n salfbnet python=3.8
conda activate salfbnet
conda install pytorch torchvision cudatoolkit=11.3 -c pytorch
pip install scikit-learn scipy tensorboard tqdm
pip install torchSummeryX

Alternativaly, you can install the envs from yml file. Before running the command, please revise the 'prefix' with your PC name.

conda env create -f environment.yml

2. Run

The running code will be released after our paper is published.

3. Datasets

• *Training and Validation sets are randomly split by this work.
• We will release our Pseudo-Saliency dataset after our paper is published.

4. Downloads

• Our pre-trained models

  It will be available soon.

• Our Pseudo-Saliency dataset (~24.2GB)

  It will be available soon.

  1. Downloading all zipped files, and using following command to restore the complete zip file:

  zip -F PseudoSaliency_avg_dataset.zip --out PseudoSaliency_avg.zip
  

  2. Then unzip the file:

  unzip PseudoSaliency_avg.zip
  

• Our testing saliency results on public datasets

  You can download our testing saliency resutls from this [link].

Performance Evaluation

1. Visulization Results

2. Testing Performance on DUT-OMRON, PASCAL-S, and TORONTO

3. Testing Performance on SALICON

4. Testing Performance on MIT300

5. Efficiency Comparison

Pseudo-Saliency Dataset

1. Annotation

2. Pseudo Saliency Distribution

Acknowledgement

• Data Collection

  We collect color images from ImageNet and SOD datasets including CSSD, ECSSD, HKU-IS, MSRA-B, MSRA10K, THUR15K.

• Pseudo-Annotators

  We use 5 models as our Pseudo-Annotators, including DeepGazeIIE, UNISAL, MSINet, EMLNet, CASNetII.

  We appriciate their public datasets and codes.