This repository is the imprimentation of "SODA: Story Oriented Dense Video Captioning Evaluation Flamework" published at ECCV 2020 pdf. SODA measures the performance of video story description systems.
v1.1 (2021/5)
- Added new option "--multi_reference" to deal with multiple reference.
SODA selects the reference that has the maximum f1 for each video, and returns macro averaged scores. - Fixed BertScore import
python 3.6+ (developed with 3.7)
- Numpy
- tqdm
- pycocoevalcap (Python3 version)
- BERTScore (optional)
You can run SODA by specifying the path of system output and that of ground truth. Both files should be the json format for ActivityNet Captions.
python soda.py -s path/to/submission.json -r path/to/ground_truth.json
You can run on the multiple reference setting, with --multi_reference
option.
python soda.py --multi_reference -s path/to/submission.json -r path/to/ground_truth1.json path/to/ground_truth2.json
You can try other sentence evaluation metrics, e.g. CIDEr and BERTScore, with -m
option.
python soda.py -s path/to/submission.json -m BERTScore
Please use the same format as ActivityNet Challenge
{
version: "VERSION 1.0",
results: {
"sample_id" : [
{
sentence: "This is a sample caption.",
timestamp: [1.23, 4.56]
},
{
sentence: "This is a sample caption 2.",
timestamp: [7.89, 19.87]
}
]
}
external_data: {
used: False,
}
}
@inproceedings{Fujita2020soda,
title={SODA: Story Oriented Dense Video Captioning Evaluation Flamework},
author={Soichiro Fujita and Tsutomu Hirao and Hidetaka Kamigaito and Manabu Okumura and Masaaki Nagata},
booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
month={August},
year={2020},
}
NTT License
According to the license, it is not allowed to create pull requests. Please feel free to send issues.