/SOTPN

Primary LanguagePython

SOTPN

Paper and codes of my SOTPN algoritm submitted to ICPR2020.

Because of the file size limit, I could't upload the weights of the backbone netowrk and SOTPN network, .npy and .tfrecord of the input embedding vectors.

Before run the codes.

  • Please create ./plt directory before run main.py. It will produce code histogram of each iteration.
  • Every wegiths file should be cotained at ./model directory. Please create ./model directory on your own.
  • Every .npy files should be contained at ./npy directory. Please creatd ./npy directory on your own.
  • Every .tfrecord files should be contaied at ./src directory.

Python file description.

  • main.py: It runs the SOTN network.
  • backbone_eval.py: It evaluates the backbone network (in here, arcface network) based on the cosine distance.
  • original_IOM.py: It evaluates the Index-of-Max (IoM) hashing.
  • permutation.py: It evaluates the permutated SOTN hash codes.

Directory description.

  • backbones: The arcface backbone network.
  • configs: Configuration of the arcface backbone network.
  • model: The weights of the arcface backbone network and the SOTN network.
  • npy: .npy files for evaluation.
  • plt: Code histogram of each iteration will be generated here when you run main.py.
  • src: Utility codes which are necessary for my codes.
  • unlinkability: Unlinkability evaluation tools of my SOTN network.