Code for CVPR 2018 paper "Hashing as Tie-Aware Learning to Rank"
This repository contains Matlab/MatConvNet implementation for the following paper:
"Hashing as Tie-Aware Learning to Rank",
Kun He, Fatih Cakir, Sarah Adel Bargal, and Stan Sclaroff.
IEEE CVPR, 2018 (arXiv)
If you use this code in your research, please cite:
@inproceedings{He_2018_TALR,
title={Hashing as Tie-Aware Learning to Rank},
author={Kun He and Fatih Cakir and Sarah Adel Bargal and Stan Sclaroff},
booktitle={The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month={June},
year={2018}
}
cachedir
under the root directory to hold experimental results./matconvnet
(for training CNNs)./vlfeat
.
Note: this is only necessary for computing the regular tie-agnostic AP metric.
We provide efficient implementation for the tie-aware metrics in +eval
.data/README.md
.startup.m
run_*.m
files.
For example, run_cifar_s1(32)
will run the Setting 1 experiment on the CIFAR-10 dataset, with 32-bit hash codes, using the default parameters therein.+demo/
with your parameter choices.
See main/get_opts.m
for the parameters.MIT License, see LICENSE
For questions/comments, feel free to contact:
apr_s_forward.m
and apr_s_backward.m
)
and tie-aware NDCG (ndcgr_s_forward.m
and ndcgr_s_backward.m
).
They attain similar performance compared to the original versions, but are much simpler to implement.
The derivations can be found in the appendix of the arxiv version of the paper (v4).