Skip to content

README.md

Latest commit

 

History

History
50 lines (46 loc) · 2.27 KB

README.md

File metadata and controls

50 lines (46 loc) · 2.27 KB

HQT: Unsupervised Hashing Network with Hyper Quantization Tree

This repository contains PyTorch-based implementation of accepted BMVC 2024 paper: Unsupervsied Hashing Network with Hyper Quantization Tree.

Unsupervised Hashing Network with Hyper Quantization Tree
Sungeun Kim, Jongbin Ryu
The British Machine Vision Conference (BMVC), 2024

Train scripts

  • HQT+BihalfNet(CIFAR10-I)
python train.py --data-path ../datasets --dataset-type cifar10 --num-train 5000 --num-query 10000 \
  --hash-model bihalf --hqt -c 16 -l 1e-4 -b 64 -m 0.9 -a vgg16 -e 300 -w 8 --weight-decay 5e-4  \
  --lr-decay 120 --gamma 6 --test-map 10
  • HQT+BihalfNet(CIFAR10-II)
python train.py --data-path ../datasets --dataset-type cifar10 --num-train 50000 --num-query 10000 \
  --hash-model bihalf --hqt -c 16 -l 1e-4 -b 64 -m 0.9 -a vgg16 -e 300 -w 8 --weight-decay 5e-4  \
  --lr-decay 120 --gamma 6 --test-map 10 --topk 1000

Validation

We provide cifar10 implementation for bihalfnet and 16bit checkpoints for protocols I and II. [LINK]

  • HQT+Bihalf(CIFAR10-I)
python evaluate.py --checkpoint ./hqt_bihalf_cifar10_I_16.pth --num-train 5000 --num-query 10000
  • HQT+BihalfNet(CIFAR10-II)
python evaluate.py --checkpoint ./hqt_bihalf_cifar10_II_16.pth --num-train 50000 --num-query 10000 --topk 1000

References

If HQT has been useful in your work, please consider citing it.

@inproceedings{Kim_2024_BMVC,
    author    = {Sungeun Kim and Jongbin Ryu},
    title     = {Unsupervised Hashing Network with Hyper Quantization Tree},
    booktitle = {35th British Machine Vision Conference 2024, {BMVC} 2024, Glasgow, UK, November 25-28, 2024},
    publisher = {BMVA},
    year      = {2024},
    url       = {https://papers.bmvc2024.org/0482.pdf}
}

Acknowledgements

We experimented with HQT by BihalfNet, CIBHash, CIMON and UHSCM our code is based on their implementation.