mmeendez8 / autoencoder

VAE implementation in tensorflow

Tensorflow VAE

This repository contains an implementation of a Variational Autoencoder. See in depth tutorial on Medium post titled VAEs I ! Generating images with Tensorflow .

Parameters

-- learning_rate      Initial learning rate
--epochs              Number of training epochs 
--batch_size          Minibatch training size
--latent_dim          Number of latent space dimensions
--test_image_number   Number of test images to recover during training
--inputs_decoder      Size of decoder input layer
--dataset             Dataset name [mnist, fashion-mnist]
--logdir              Log folder
--plot_latent         Plot latent space distribution [True, False]

Run

$ python autoencoder.py  --latent_dim=2 --dataset=mnist --epochs=50

Example

autoencoder's Issues

Strange Problem

Hi, guys,

I have a problem with the following code:

https://github.com/mmeendez8/Autoencoder/blob/master/autoencoder.py.

The code is now running on my computer and no error messages are displayed. But the result is strange as you can see on the pictures. Once the code is finished, the results are displayed in a folder. But only the results of the 0 epoch are shown. After that nothing is displayed. I did not change anything in the code and just copied and pasted it. Can somebody help me and tell me what the problem is ?

Furthermore only the numbers are shown but not the results of the fashion mnist.

An error when feeding input samples

There is an error in your implementation: Can 't feed input_holder (shape: [2, 14, 14]) into z (shape: (?, 10):

Traceback (most recent call last): File "olala.py", line 185, in <module> artificial_image = sess.run(output, feed_dict={z: input_holder}) File "/home/theanh/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 929, in run run_metadata_ptr) File "/home/theanh/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1128, in _run str(subfeed_t.get_shape()))) ValueError: Cannot feed value of shape (1, 2) for Tensor 'Encoder/add:0', which has shape '(?, 10)'

my result is very bad

the reconstruction loss is very high as you can see in the image, but I don't know why