Multi-label Image Recognition with Partial Labels

Implementation of papers:

• Structured Semantic Transfer for Multi-Label Recognition with Partial Labels
  36th Association for the Advance of Artificial Intelligence (AAAI), 2022.
  Tianshui Chen*, Tao Pu*, Hefeng Wu, Yuan Xie, Liang Lin. (* equally contributed)
• Semantic-Aware Representation Blending for Multi-Label Image Recognition with Partial Labels
  36th Association for the Advance of Artificial Intelligence (AAAI), 2022.
  Tao Pu*, Tianshui Chen*, Hefeng Wu, Liang Lin. (* equally contributed)
• Heterogeneous Semantic Transfer for Multi-label Recognition with Partial Labels
  International Journal of Computer Vision (IJCV), 2024.
  Tianshui Chen*, Tao Pu*, Lingbo Liu, Yukai Shi, Zhijing Yang, Liang Lin. (* equally contributed)
• Dual-Perspective Semantic-Aware Representation Blending for Multi-Label Image Recognition with Partial Labels
  Expert Systems with Applications (ESWA), 2024.
  Tao Pu, Tianshui Chen, Hefeng Wu, Yukai Shi, Zhijing Yang, Liang Lin.

Preliminary

1. Donwload data.zip ([One Drive] [Baidu Drive]), and unzip it.
2. Modify the lines 16-19 in config.py.
3. Create servel folders (i.e., "exp/log", "exp/code", "exp/checkpoint", "exp/summary") to record experiment details.

Usage

cd HCP-MLR-PL

# modify experiment settings
# range of <#model>: SST, SARB, HST, DSRB
vim scripts/<#model>.sh

./<#model>.sh

Common Issues

1. How to generate the partial labels?

Since all the datasets have complete labels, we randomly drop a certain proportion of positive and negative labels to create partially annotated datasets. To control the remaining labels' proportion, we can modify the variable 'prob' in each file of the directory 'scripts'. Specifically, we provide the partial labels generating function in 'datasets/coco2014.py', 'datasets/vg.py', 'datasets/voc2007.py'.

As you can find, in each dataset class, we provide two elements of annotations: (1) 'labels': original ground truth annotations whose shape is $N * C$; (2) 'changeLabels': generated partial labels whose shape is $N * C$. For ease of reproducibility, we freeze the random seed of generating partial labels.

Notes: for convenience, we also provide partial labels of each dataset on all known label proportions. ([One Drive] [Baidu Drive])

Citation

@inproceedings{Chen2022SST,
  author={Chen, Tianshui and Pu, Tao and Wu, Hefeng and Xie, Yuan and Lin, Liang},
  title={Structured semantic transfer for multi-label recognition with partial labels},
  booktitle={Proceedings of the AAAI conference on artificial intelligence},
  year={2022},
  volume={36},
  number={1},
  pages={339--346}
}

@inproceedings{Pu2022SARB,
  author={Pu, Tao and Chen, Tianshui and Wu, Hefeng and Lin, Liang},
  title={Semantic-aware representation blending for multi-label image recognition with partial labels},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2022},
  volume={36},
  number={2},
  pages={2091--2098}
}

@article{Chen2024HST,
  title={Heterogeneous Semantic Transfer for Multi-label Recognition with Partial Labels},
  author={Chen, Tianshui and Pu, Tao and Liu, Lingbo and Shi, Yukai and Yang, Zhijing and Lin, Liang},
  journal={International Journal of Computer Vision},
  year={2024},
  issn={1573-1405},
  doi={10.1007/s11263-024-02127-2},
}

@article{Pu2024DSRB,
  title = {Dual-perspective semantic-aware representation blending for multi-label image recognition with partial labels},
  author = {Tao Pu and Tianshui Chen and Hefeng Wu and Yukai Shi and Zhijing Yang and Liang Lin}
  journal = {Expert Systems with Applications},
  year = {2024},
  pages = {123526},
  issn = {0957-4174},
  doi = {https://doi.org/10.1016/j.eswa.2024.123526},
}

Contributors

For any questions, feel free to open an issue or contact us:

• [email protected]
• [email protected]