IANet

Official code repository for paper Local-Global Interaction and Progressive Aggregation for Video Salient Object Detection (ICONIP 2022)

Overall architecture of the proposed IANet.

Usage

Each dataset corresponds to a txt path file, with each row arranged by img_path, gt_path and flow_path.

Training

1. Download the training dataset (containing DAVIS16, DAVSOD and DUTS-TR) from Baidu Driver (PSW:wuqv).
2. Download the pre_trained ResNet34 backbone to your specified folder.
3. The training of entire model is implemented on a NVIDIA TITAN X (Pascal) GPU:

• Run python main.py --mode=train

Testing

1. Download the test dataset (containing DAVIS16, DAVSOD, FBMS, SegTrack-V2, VOS and ViSal) from Baidu Driver (PSW:wuqv).
2. Download the final trained model from Baidu Driver (PSW:u9wa).
3. Run python main.py --mode=test.

Result

1. The saliency maps can be download from Baidu Driver (PSW: u76y).
2. Evaluation Toolbox: We use the standard evaluation toolbox from DAVSOD benchmark.

Quantitative comparisons with SOTA methods on five public VSOD datasets in term of three evaluation metrics. The best results are highlighted in bold.

Citation

Please cite the following paper if you use this repository in your research:

@article{min2022ianet,
  title={Local-Global Interaction and Progressive Aggregation for Video Salient Object Detection},
  author={Min, Dingyao and Zhang, Chao and Lu, Yukang and Fu, Keren and Zhao, Qijun},
  booktitle={The International Conference on Neural Information Processing},
  year={2022}
}