Code for the CVPR2025 paper "Attribute-formed Class-specific Concept Space: Endowing Language Bottleneck Model with Better Interpretability and Scalability"

Set up environments

We run our experiments using Python 3.9.19 . You can install the required packages using:

conda create --name ALBM python=3.9.19
conda activate ALBM
pip install -r requirements.txt

Directories

• clip/ saves the code of CLIP model
• configs/ saves the config files for all experiments. You can modify the config files to change the system arguments
• dataset/ saves the code for building the dataset
• docs/ saves instructions on how to train and test
• scripts/ saves the script files for training and testing
• trainer/ saves the models

The results will be saved in output/

Training and Testing

The relevant script files are saved in Run,md

Please cite our paper if you find it useful!

@misc{zhang2025attributeformedclassspecificconceptspace,
      title={Attribute-formed Class-specific Concept Space: Endowing Language Bottleneck Model with Better Interpretability and Scalability}, 
      author={Jianyang Zhang and Qianli Luo and Guowu Yang and Wenjing Yang and Weide Liu and Guosheng Lin and Fengmao Lv},
      year={2025},
      eprint={2503.20301},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.20301}, 
}