Code for "Re-thinking Co-Salient Object Detection" IEEE TPAMI2021.
The caffe package is borrowed from https://github.com/BVLC/caffe
- VGG16 model on ImageNet:
models/deploy_vgg16CAM.prototxt
weights:[http://cnnlocalization.csail.mit.edu/demoCAM/models/vgg16CAM_train_iter_90000.caffemodel]
- simple run ./stage1/demo.m, then we can obtain the initial cosal activations
- Python 3.7, PyTorch 1.1.0
- python ./stage2/run_sample.py (Thanks for Ahn et al.'s implement, our postprocessing is inspired by IRNET)
Baidu Cloud: https://pan.baidu.com/s/19hIlViLbby-a7vQw17ZTVw Fetchcode: f4p3
Google Cloud: https://drive.google.com/file/d/1AK9UNR5mLHQakOTkRawcKSDg8YwIu-xJ/view?usp=sharing
The sub-net of Co-EGNet is our previous version EGNet. Please refer to the training details in: https://github.com/JXingZhao/EGNet
CoSOD3K dataset: https://github.com/DengPingFan/CoSOD3K
If you use this code, please cite our paper:
@article{deng2021re,
title={Re-thinking co-salient object detection},
author={Deng-Ping, Fan and Tengpeng, Li and Zheng, Lin and Ge-Peng, Ji and Dingwen, Zhang and Ming-Ming, Cheng and Huazhu, Fu and Jianbing, Shen},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
doi={10.1109/TPAMI.2021.3060412},
year={2021}
}