Awesome

CTDNet

The PyTorch code for ACM MM2021 paper "Complementary Trilateral Decoder for Fast and Accurate Salient Object Detection"

Requirements

• Python 3.6
• Pytorch 1.4+
• OpenCV 4.0
• Numpy
• TensorboardX
• Apex

Dataset

Download the SOD datasets and unzip them into data folder.

Train

cd src
python train.py

• We implement our method by PyTorch and conduct experiments on a NVIDIA 1080Ti GPU.
• We adopt pre-trained ResNet-18 and ResNet-50 as backbone networks, which are saved in res folder.
• We train our method on DUTS-TR and test our method on other datasets.
• After training, the trained models will be saved in out folder.

Test

cd src
python test.py

• After testing, saliency maps will be saved in eval folder.

Results

• CTDNet-18： saliency maps (提取码：b6ba)； trained model (提取码：ftmz）
• CTDNet-50： saliency maps (提取码：j1zq)； trained model (提取码：ehvv）

Evaluation

    cd eval
    matlab main

• We use MATLAB code to evaluate the performace of our method.

Citation

• If you find this work is helpful, please cite our paper

@inproceedings{zhao2021complementary,
  title={Complementary Trilateral Decoder for Fast and Accurate Salient Object Detection},
  author={Zhao, Zhirui and Xia, Changqun and Xie, Chenxi and Li, Jia},
  booktitle={Proceedings of the 29th ACM International Conference on Multimedia},
  pages={4967--4975},
  year={2021}
}

Reference

This project is based on the implementation of F3Net.