- Python 3.x (tested with Python 3.6)
- TensorFlow v1.x (tested with 1.8)
- tqdm (for the progress bar)
Just run this command
python main.py
Open the MNIST Plot.ipynb with Jupyter Notebook/Lab.
A pretrained model for MNIST is included in the repository here.
Please download the zip file and decompress it on assets/pretrained_models/celeba/last*. Or, you can easily modify a path at the first cell on the notebook.
- Learning statistics
- Reconstruction Results
(top): original images from MNIST validation set, (bottom): reconstructed image - As can be seen in the MNIST jupyter notebook, our novel implementation of quasi-euclidean metric gives a similarity score of 1 to WAE-MMD synthesized images and the ground truth (original) image. It is justifiable since WAE-MMD do a very well job at synthesizing MNIST data especially since the images only have a few amount of pixels as well as not a lot of features that the model has to recognize.
- Random Sampled Images
