DMMCS: A Data-driven Guided Decoding Mechanism for Diagnostic Captioning

Distance from Median Maximum Cosine Similarity (DMMCS)

This repository contains the official codebase for DMMCS, our novel data-driven guided decoding algorithm featured in ACL Findings 2024. You can find our paper "A data-driven guided decoding mechanism for Diagnostic Captioning" here. DMMCS stands for Distance from Median Maximum Cosine Similarity.

Installation

To get started with our framework, follow these steps to clone the repository and install the required packages. We recommend using a virtual environment for package installation to ensure a clean and isolated setup.

Step 1: Clone the repository

git clone https://github.com/nlpaueb/dmmcs.git
cd dmmcs

Step 2: Create and activate a virtual environment

We have tested our framework for both Conda and Virtualenv environments.

Conda

conda create -n dmmcs_venv python=3.9
conda activate dmmcs_venv
pip install -r requirements.txt

Virtualenv

virtualenv dmmcs_venv
source dmmcs_venv/bin/activate
pip install -r requirements.txt

Usage

Step 1: Calculate your data-specific stats

First, you need to download the en_core_web_sm package from the spacy library.

python -m spacy download en_core_web_sm

Then, you have to run the stats_extraction.py script.

python3 utils/stats_extraction.py --config config/stats_extractor_config.json

Please make sure to adjust the config/stats_extractor_config.json in order to match your local file directories.

This script generates four files that will be needed for our guided-decoding mechanism. These files can be found under the snapshots/artifacts directory.

Step 2: Run training and/or inference

You can train and/or infer from an InstructBLIP model using the proposed guided-decoding mechanism with:

python3 instructBLIP-ft.py --config ../config/config.json

Please make sure to adjust the config/config.json args file to your own local paths and directories.

Set the do_dmmcs option equal to True in order to use the dmmcs guided-decoding mechanism during inference instead of the vanilla beam search.

Licence

This repository is licensed under the MIT license. See LICENSE for more details.

Contact

For any questions, inquiries or suggestions, please feel free to reach out at pkaliosis@aueb.gr and/or annis@aueb.gr.

Citation

If you would like to use our work, please cite us using the following bibtex reference:

@inproceedings{kaliosis-etal-2024-data,
    title = "A Data-Driven Guided Decoding Mechanism for Diagnostic Captioning",
    author = "Kaliosis, Panagiotis  and
      Pavlopoulos, John  and
      Charalampakos, Foivos  and
      Moschovis, Georgios  and
      Androutsopoulos, Ion",
    editor = "Ku, Lun-Wei  and
      Martins, Andre  and
      Srikumar, Vivek",
    booktitle = "Findings of the Association for Computational Linguistics ACL 2024",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand and virtual meeting",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.findings-acl.444",
    pages = "7450--7466",
}