Skip to content

Repository files navigation

Improved BERTScore for image captioning evaluation

The code for improved BERTScore evaluation of our paper, Switching to Discriminative Image Captioning by Relieving a Bottleneck of Reinforcement Learning (WACV 2023).

Acknowledgment

The code is based on improved-bertscore-for-image-captioning-evaluation. We thank the authors of the repository.

Setup

git clone https://github.com/ukyh/bertspp_cocout.git
cd bertspp_cocout

conda create --name bertspp python=3.6
conda activate bertspp

pip install -r requirements.txt

# Test run
python -u run_metric_custom.py --file samples --dir example	

Downloads

mkdir data; cd data
wget http://cs.stanford.edu/people/karpathy/deepimagesent/caption_datasets.zip
unzip caption_datasets.zip
rm -f caption_datasets.zip

Then, download cocotalk_disc_text.zip and unzip it into data/:
unzip cocotalk_disc_text.zip -d data/

Run

Copy the output files to evaluate from switch_disc_caption (the files under eval_results).
Then, run the following commands.

cd bertspp_cocout
conda activate bertspp
export PYTHONPATH=$PYTHONPATH:`pwd`

# NOTE: the end of the file name has to be "_val.json" or "_test.json"
ID=sample_test
python -u proc_bert_score_pp.py --hyp eval_results/${ID}.json
python -u run_metric_custom.py --file ${ID} --dir coco_caps

Reference

If you find the paper or this code useful, please consider citing:

@inproceedings{honda2023switch,
  title={Switching to Discriminative Image Captioning by Relieving a Bottleneck of Reinforcement Learning},
  author={Honda, Ukyo and Taro, Watanabe and Yuji, Matsumoto},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year={2023}
}

@inproceedings{yi2020improving,
  title={Improving image captioning evaluation by considering inter references variance},
  author={Yi, Yanzhi and Deng, Hangyu and Hu, Jinglu},
  booktitle={ACL},
  year={2020}
}

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages