wyj2046/deephash

An Open-Source Package for Deep Learning to Hash (DeepHash)

DeepHash

DeepHash is a lightweight deep learning to hash library that implements state-of-the-art deep hashing/quantization algorithms. We will implement more representative deep hashing models continuously according to our released deep hashing paper list. Specifically, we welcome other researchers to contribute deep hashing models into this toolkit based on our framework. We will announce the contribution in this project.

The implemented models include:

• DQN: Deep Quantization Network for Efficient Image Retrieval, Yue Cao, Mingsheng Long, Jianmin Wang, Han Zhu, Qingfu Wen, AAAI Conference on Artificial Intelligence (AAAI), 2016
• DHN: Deep Hashing Network for Efficient Similarity Retrieval, Han Zhu, Mingsheng Long, Jianmin Wang, Yue Cao, AAAI Conference on Artificial Intelligence (AAAI), 2016
• DVSQ: Deep Visual-Semantic Quantization for Efficient Image Retrieval, Yue Cao, Mingsheng Long, Jianmin Wang, Shichen Liu, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017
• DCH: Deep Cauchy Hashing for Hamming Space Retrieval, Yue Cao, Mingsheng Long, Bin Liu, Jianmin Wang, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018

Requirements

• Python3
• Other packages: pip install tensorflow-gpu scipy h5py

To import the algorithms implemented in ./DeepHash, we need to add the path of ./DeepHash to environment variables as:

export PYTHONPATH=/path/to/project/DeepHash/DeepHash:$PYTHONPATH

Data Preparation

In data/cifar10/train.txt, we give an example to show how to prepare image training data. In data/cifar10/test.txt and data/cifar10/database.txt, the list of testing and database images could be processed during predicting procedure. If you want to add other datasets as the input, you need to prepare train.txt, test.txt and database.txt as CIFAR-10 dataset.

Get Started

Pre-trained model

You should manually download the model file of the Imagenet pre-tained AlexNet from here or from release page and unzip it to /path/to/project/DeepHash/architecture/pretrained_model.

Make sure the tree of /path/to/project/DeepHash/architecture looks like this:

├── __init__.py
├── pretrained_model
       └── reference_pretrain.npy

Training and Testing

The example of $method (DCH, DVSQ, DQN and DHN) can be run with the following command:

cd example/$method/
python train_val_script.py --gpus "0,1" --"other parameters descirbe in train_val_script.py"

Citations

If you find DeepHash is useful for your research, please consider citing the following papers:

@InProceedings{cite:AAAI16DQN,
  Author = {Yue Cao and Mingsheng Long and Jianmin Wang and Han Zhu and Qingfu Wen},
  Publisher = {AAAI},
  Title = {Deep Quantization Network for Efficient Image Retrieval},
  Year = {2016}
}

@InProceedings{cite:AAAI16DHN,
  Author = {Han Zhu and Mingsheng Long and Jianmin Wang and Yue Cao},
  Publisher = {AAAI},
  Title = {Deep Hashing Network for Efficient Similarity Retrieval},
  Year = {2016}
}

@InProceedings{cite:CVPR17DVSQ,
  Title={Deep visual-semantic quantization for efficient image retrieval},
  Author={Cao, Yue and Long, Mingsheng and Wang, Jianmin and Liu, Shichen},
  Booktitle={CVPR},
  Year={2017}
}

@InProceedings{cite:CVPR18DCH,
  Title={Deep Cauchy Hashing for Hamming Space Retrieval},
  Author={Cao, Yue and Long, Mingsheng and Bin, Liu and Wang, Jianmin},
  Booktitle={CVPR},
  Year={2018}
}

Contacts

Maintainers of this library:

• Yue Cao, Email: [email protected]
• Bin Liu, Email: [email protected]