LICO: Explainable Models with Language-Image COnsistency (NeurIPS 2023)

Abstract & Framework

Framework of LICO is shown in the following figure:

Results

Qualitative and quantitative results obtained by baselines and their LICO versions.

Response to Reproducibility Study at TMLR

We recently read a reproduction study of our paper and found that the team encountered challenges in reproducing our results, which led to contrasting conclusions. We have carefully read their paper and code and found coding errors and misunderstandings of our methods. Specifically, this includes erroneous implementation of Text Features and data processing, which was inconsistent with our methods, such as incorrect dimensions of text features and prompt tokens. In addition, we found some other settings that affected the results. Detailed information can be found in this attachment: Response to Reproducibility of LICO.pdf

We hope our report can help in addressing these issues and correcting erroneous conclusions. We also apologize for not having updated our code in time and we have now updated our code.

References

If you find the code useful for your research, please consider citing

@inproceedings{lei2023lico,
  title={LICO: Explainable Models with Language-Image COnsistency},
  author={Lei, Yiming and Li, Zilong and Li, Yangyang and Zhang, Junping and Shan, Hongming},
  booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
  year={2023}
}