DBCR - Dual-Branch Compositional Reasoning

Official Repository for:

Chengtai Li, Guosheng Su, Jianfeng Ren, Ruibin Bai, Yitian Zhao, Xudong Jiang. "DBCR: Exploiting Both Intra-cluster and Extra-cluster Relations for Compositional Reasoning". ICASSP 2025.

Dataset

• CVR (NeurIPS 2022)
• MC2R (IJCAI 2024)

Dependencies

• numpy: 1.24.4
• pandas: 1.4.2
• pillow: 9.0.1
• python: 3.8.0
• pytorch-lightning: 1.6.0
• pyyaml: 6.0.1
• scikit-learn: 1.0.2
• scipy: 1.8.0
• tensorboard: 2.8.0
• timm: 0.5.4
• torch: 2.0.1
• torchvision: 0.15.2

Pretrained Models

The pre-trained ResNet50 can be downloaded from here.

Code

To reproduce the results,

• In DBCR/jobs/train_array.sh, set checkpoint to path_to_pretrained_model.
• In DBCR/jobs/train.sh, set data_dir as path_to_dataset, and set WORK_PATH as path_to_experiments.
• Run the following command

$ ./jobs/train_array.sh

Citation

@inproceedings{li2025dbcr, 
    title={DBCR: Exploiting Both Intra-cluster and Extra-cluster Relations for Compositional Reasoning}, 
    author={Li, Chengtai and Su, Guosheng and Ren, Jianfeng and Bai, Ruibin and Zhao, Yitian and Jiang, Xudong},
    booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
    year={2025}
}

Acknowledgement

We thank the authors of CVR and MC2R for their valuable prior work.