The official repo of the ECCV 2020 paper Highly Efficient Salient Object Detection with 100K Parameters.
CSNet will be released in the next few days.
Cross Stage Fusion (CSF) part + Res2Net
is in the CSF+Res2Net
subfolder.
CSF can be integrated into ImageNet pre-trained large models such as Res2Net to achieve SOTA performance with less parameters.
If you find this work or code is helpful in your research, please cite:
@InProceedings{gao2020sod100k,
author = {Gao, Shang-Hua and Tan, Yong-Qiang and Cheng, Ming-Ming and Lu, Chengze and Chen, Yunpeng and Yan, Shuicheng},
title = {Highly Efficient Salient Object Detection with 100K Parameters},
booktitle = {ECCV},
year = {2020},
}
@article{gao2019res2net,
title={Res2Net: A New Multi-scale Backbone Architecture},
author={Gao, Shang-Hua and Cheng, Ming-Ming and Zhao, Kai and Zhang, Xin-Yu and Yang, Ming-Hsuan and Torr, Philip},
journal={IEEE TPAMI},
year={2020},
doi={10.1109/TPAMI.2019.2938758},
}