Code for selecting informative labels of images based on a known distribution of labels.
See our CVPR 2019 Paper
If you find IOTA useful in your research, please consider citing:
@inproceedings{bracha2019informative,
title={Informative Object Annotations: Tell Me Something I Don't Know},
author={Bracha, Lior and Chechik, Gal},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages={12507--12515},
year={2019}
}
git clone ssh://git@github.com:liorbracha/iota.git
$ conda install --yes --file requirements.txt
to install in a “single package mode”:
$ while read requirement; do conda install --yes $requirement; done < requirements.txt
wget https://chechiklab.biu.ac.il/~brachalior/IOTA/data/iota10K/iota_raw.csv.tar.gz
Download Image-Level Labels from the Open Images Dataset to data/oid
To change the default location of the data or results directory:
export RES_DIR=/where/results/are/written
export OID_DIR=/where/data/is/located/
# data
# |_ ground_truth
# |_oid
# |_classes
# |_train
# |_validation
# |_test
# results
# |_models
# |_counts