README.md

Deep Adversarial Quantization Network for Cross-modal Retrieval

1. Environment

This is a demo on MIRFlickr dataset for our paper. We finish experiments on a server with one NVIDIA GeForce 1080Ti GPU.

The python version is 3.5.2. These python packages are used for our experiments:

   -opencv-python          3.4.2.16
   -tensorflow-gpu         1.4.0 
   -numpy                  1.16.2  

You can directly run the python file train_scripts.py to get results after relevant data prepared.

2. Relevant data and setup

Please preprocess dataset to appropriate the input format and modify the partition of dataset in data_handler.py.

Or you can download the data we preprocessed from the pan.baidu.com.

    a. modify the path of pre-trained VGGnet weights file in bone_net.py 
    link: https://pan.baidu.com/s/1vag9Cag40zAxySMKt0i9lg  
    password: ij9g  
    
    b. modify the dataset path and partition in data_handler.py   
    MIR-FLICKR link: https://pan.baidu.com/s/1ea-TvNZAcG4e6IWZRzeB9Q  
    password: a1cv

3. Thanks

We should thank to these kind researchers, who unreservedly share their source code and advice to us.

Including but not limited to these:

    1. QingYuan Jiang,  Nanjing          University,  P.R. China
    2. Mingsheng Long,  Tsinghua         University,  P.R. China
    3. Yue        Cao,  Tsinghua         University,  P.R. China
    4. Zhikai      Hu,  HongKong Baptist University,  P.R. China

4. Contact

If you have any question, don't be hesitate to contact Yu Zhou at 18990848997@163.com.

If you are a Chinese, you can surely write an e-mail in Chinese.

If you find DAQN useful in your research, please consider citing us:

---

Yu Zhou, Yong Feng, Mingliang Zhou, Baohua Qiang, Leong Hou U and Jiajie Zhu, "Deep Adversarial Quantization Network for Cross-Modal Retrieval," ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021, pp. 4325-4329, doi: 10.1109/ICASSP39728.2021.9414247.

---