adversarial-discriminative-domain-adaptation-for-hashing

An implementation of "adversarial discriminative domain adaptation". The original work targets classification task, but this project implements it for hashing purpose. The classification loss is replaced with pairwise similarity loss

How to run

Make sure pytorch, and the latest version of ml_toolkit are installed Then run run.py

Test results

Training on MNIST(mini), MNIST_M(10% mini), Testing on MNIST_M

• Before domain adaptation

  • Precision@2 = 39.83%
  • MAP@2 = 42.84%

• After domain adaptation

  • Precision@2 = 68.00%
  • MAP@2 = 70.36%