To download the framework: saliency framework.pdf
ArXiv: https://arxiv.org/abs/2204.06788
You need to install Pytorch (preferred 1.7.1) and some basic libraries including PIL, cv2, numpy, etc.
The datasets can be downloaded from the following links. We follow the training and testing data similar to the previous methods.
Option 1: Download from Google drive of an existing SOD Method (UCNet): Click here
Option 2: Download from GitHub of existing SOD Method (D3Net): Click here
Option 3: Follow a dedicated GitHub page for SOD datasets: Click here
Run train.py (current training code supports RGB-D dataset training, you can request for RGB as well or tune the code yourself)
In the training code, there is function call to the testing code which then evaluates the model's performance and stores the results in corresponding folders.
Notice: Please follow the paper pdf to view the references.
Please cite our following papers on Saliency Detection if you like our work:
@InProceedings{hussain2022Pyramidal,
author = {Hussain, Tanveer and Anwar, Abbas and Anwar, Saeed and Petersson, Lars and Baik, Sung Wook},
title = {Pyramidal Attention for Saliency Detection},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2022},
pages = {2878-2888}
}
}
@misc{hussain2021densely,
title={Densely Deformable Efficient Salient Object Detection Network},
author={Tanveer Hussain and Saeed Anwar and Amin Ullah and Khan Muhammad and Sung Wook Baik},
year={2021},
eprint={2102.06407},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
Some of the functions in codes are inspired by UCNet GitHub repository. The authors are thankful for their nice and explained SOD GitHub page.
I would be happy to guide and assist in case of any questions and I am open to research discussions and collaboration in Saliency Detection domain. Ping me at htanveer3797 [at] [gmail] [.com]