By Josiah Wang
This repository contains the libraries and scripts used for evaluating phrase localization in my ICCV 2019 paper (Phrase Localization Without Paired Training Examples).
I found existing evaluation scripts for the phrase localization task difficult to setup and use. I only really needed to evaluate my output, not to run other people's models so that I can evaluate my output! Thus, this toolkit was born!
I wrote it in standard Python, so the script does not need any other dependencies. You only need to provide the script a list of ground truth bounding boxes and a list of predicted bounding boxes, and it will return the accuracy. I hope having a simple and standard evaluation script for the task will make life easier for everyone!
The toolkit is a Python module located in the lib/ directory. Please refer to the doc comments in the code for explanations and usage.
Ground truth annotations for various datasets are provided in the data/ directory.
An example script is available as demo.py.
If you use this evaluation toolkit, please cite the following work:
Josiah Wang and Lucia Specia (2019). Phrase Localization Without Paired Training Examples. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV).
@InProceedings{WangSpecia:2019,
author = {Wang, Josiah and Specia, Lucia},
title = {Phrase Localization Without Paired Training Examples},
booktitle = {Proceedings of the IEEE/CVF Internaitonal Conference on Computer Vision (ICCV)},
publisher = {{IEEE}},
month = oct,
year = {2019},
pages = {},
address = {Seoul, South Korea}
}
GNU General Public License v3.0