HiHPQ

This repo includes the official implementation of HiHPQ: Hierarchical Hyperbolic Product Quantization for Unsupervised Image Retrieval.

Reproduction

Codes are tested on the following python environment.

python  3.9.13
torch  1.12.1
scikit-learn  1.0.2
numpy 1.26.4

Please refer to the ./scripts/ directory to reproduce the main results. An example run is:

sh ./scripts/cifar10_ii/16bits.sh

Note that for experiments on NUS-WIDE and Flickr, you should first download the raw datasets. You can find the download information in this page.

Acknowledgements

The code implementation is based on the MeCoQ, HyboNet and HCNN.

Reference

If you find this code useful in your research, please cite the following paper:

@inproceedings{qiu2024hihpq,
  title={HiHPQ: Hierarchical Hyperbolic Product Quantization for Unsupervised Image Retrieval},
  author={Qiu, Zexuan and Liu, Jiahong and Chen, Yankai and King, Irwin},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2024}
}