Code for BCR

Code for "Leveraging Bilateral Correlations for Multi-Label Few-Shot Learning" in IEEE Transactions on Neural Networks and Learning Systems.

If you use the code in this repo for your work, please cite the following bib entries:

@ARTICLE{TNNLS.2024.3388094,
  author={An, Yuexuan and Xue, Hui and Zhao, Xingyu and Xu, Ning and Fang, Pengfei and Geng Xin},
  journal={IEEE Transactions on Neural Networks and Learning Systems}, 
  title={Leveraging Bilateral Correlations for Multi-Label Few-Shot Learning}, 
  year={2024},
  doi={10.1109/TNNLS.2024.3388094}
}

Requirements

• Python >= 3.6
• PyTorch (GPU version) >= 1.5
• NumPy >= 1.13.3
• Scikit-learn >= 0.20

Getting started

Download the MS-COCO, CUB-200-2011, NUS-WIDE, and Visual Genome datasets.

You can change the path of the datasets in data/dataset.py.

Running the scripts

To train and test the BCR model in the terminal, use:

$ python run_bcr.py --dataset_name VG --algorithm bcr --model_name Conv4 --n_way 10 --n_shot 1 --max_epoch 500 --hidden_dim 100 --eta 0.5 --gamma 0.5 --num_workers 8 --device cuda:0 --seed 0

Acknowledgment

Our project references the code in the following repo.

CloserLookFewShot