Authors: Lior Bracha*, Eitan Shaar*, Aviv Shamsian, Ethan Fetaya, Gal Chechik
[2023/05/06] Added controls for sentences lengths
[2023/05/06] DisCLIP is publicly released!
Run pip install -r requirements.txt
or follow instructions in env.md
- Download refexp annotations and save under
data/anns
- Download the refCOCO/g/+ images and unzip it (or create symlink) to
data/images/refcoco
- Download the refclef images and unzip it (or create symlink) to
data/images/refclef
- Download the refGTA images and unzip it (or create symlink) to
data/images/refgta
Create a symlink from image storage location to data/images/{dataset}
e.g.
ln -s /path/to/coco/images/ data/images/refcoco
python run.py --dataset refcoco+ --split testa --speaker disclip --listener reclip
If you find our paper and resources useful, please cite our paper.
@article{bracha2023disclip,
title={DisCLIP: Open-Vocabulary Referring Expression Generation},
author={Bracha, Lior and Shaar, Eitan and Shamsian, Aviv and Fetaya, Ethan and Chechik, Gal},
journal={arXiv preprint arXiv:2305.19108},
year={2023}
}