pectinid16 / ISDH-Tensorflow

ISDH-Tensorflow

Implemement of Instance Similarity Deep Hashing for Multi-Label Image Retrieval

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/ISDH-48b/.

• Test:
  Run sh test-flickr.sh, and generated hash codes will be saved in ./results/ISDH_48b_test_flickr.txt.

Other Datasets:

• NUS-WIDE
• VOC2012

Acknowledgement

• Tensorflow Implementation of AlexNet