Skip to content

ghif/drcn

Repository files navigation

Deep Reconstruction-Classification Networks for Unsupervised Domain Adaptation (DRCN)

This code is an implementation of the DRCN algorithm presented in [1].

[1] M. Ghifary, W. B. Kleijn, M. Zhang, D. Balduzzi, and W. Li. "Deep Reconstruction-Classification Networks for Unsupervised Domain Adaptation (DRCN)", European Conference on Computer Vision (ECCV), 2016

Contact:

Muhammad Ghifary ([email protected])

Requirements

  • Python 2.7
  • Tensorflow-1.0.1
  • Keras-2.0.0
  • numpy
  • h5py

Usage

To run the experiment with the (grayscaled) SVHN dataset as the source domain and the MNIST dataset as the target domain

python main_sm.py

The core algorithm is implemented in drcn.py. Data augmentation and denoising strategies are included as well.

Results

The source to target reconstruction below (SVHN as the source) indicates the successful training of DRCN.

python reconstruct_images.py

alt text

The classification accuracies of one DRCN run are plotted as follows -- the results may vary due to the randomness:

python plot_results.py

alt text

Releases

No releases published

Packages

No packages published

Languages