nnizhang / CADC

source codes of "Summarize and Search: Learning Consensus-aware Dynamic Convolution for Co-Saliency Detection" (ICCV2021)

CADC

source codes of "Summarize and Search: Learning Consensus-aware Dynamic Convolution for Co-Saliency Detection" by Ni Zhang, Nian Liu, Junwei Han, and Ling Shao.

created by Ni Zhang, email: [e-mail]

Requirement

1. Pytorch 1.6.0
2. Torchvision 0.3.0
3. apex

Training

1. Download the pretrained vgg model [baidu pan fetch code: dyt4 | Google drive]. Then create pretrained_model/ folder and put the downloaded model in it.
2. Download training images, including original DUTS Class [baidu pan fetch code: 6jkx | Google drive], COCO9213 [baidu pan fetch code: 5183| Google drive], and our synthesis data [baidu pan fetch code:shyw]. Then put them in Data/ folder.

Testing

1. Download test datasets, including CoCA, CoSOD3k, CoSal150, and MSRC. Put them in Data/ folder.
2. Modify lines 54-57 and make sure the paths of test images are corrected.
3. Run test.py and the predictions will be generated in Preds/ folder.

Evaluation

We use evaluation tool from the project.

Testing on Our Pretrained CADC Model

1. Download our final model CADC.pth [baidu pan fetch code: 6sae| Google drive]. Then create checkpoint/ folder and put CADC.pth in it.
2. Comment lines 47 and 48 and uncomment lines 50 and 51 in parameter.py.
3. Run test.py and the predictions will be generated in Preds/ folder.

Our saliency maps can be download from [baidu pan fetch code: i59u | Google drive].

Citation

If you think our work is helpful, please cite

@InProceedings{Zhang_2021_ICCV,
    author    = {Zhang, Ni and Han, Junwei and Liu, Nian and Shao, Ling},
    title     = {Summarize and Search: Learning Consensus-Aware Dynamic Convolution for Co-Saliency Detection},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2021},
    pages     = {4167-4176}
}