pectinid16 / IDHN

Implementation of IDHN in TensorFlow

IDHN

Code for the following paper:

Z. Zhang and Q. Zou and Y. Lin and L. Chen and S. Wang, "Improved Deep Hashing with Soft Pairwise Similarity for Multi-label Image Retrieval", IEEE Transactions on Multimedia, 2019.

Requirements

• Linux
• Tensorflow
• NVIDIA GPU + CUDA CuDNN

Getting Started:

• Prepare the datasets
  Download the flickr dataset and put the images into folder /data/flickr/images/

• Transform the train images in tfrecord format
  Run python tf_record.py, and train-flickr.tfrecords will be generated

• Prepare the AlexNet weights trained on ImageNet
  Download from here and put it on current directory

• Train:
  Run sh train-flickr.sh, and the trained model will be saved in models/IDHN-48b/.

• Test:
  Run sh test-flickr.sh, and generated hash codes will be saved in ./results/IDHN_48b_test_flickr.txt.

Citation

Please cite our paper if you use this code in your own work:

@article{zhang2019tmm,
   author = {Zhang, Zheng and Zou, Qin and Lin, Yuewei and Chen, Long and Wang, Song},
   title = {Improved deep hashing with soft pairwise similarity for multi-label image retrieval},
   journal = {IEEE Transactions on Multimedia},
   year = {2019}
}