lmescheder/adversarialvariationalbayes

This repository contains the code to reproduce the core results from the paper "Adversarial Variational Bayes: Unifying Variational Autoencoders and Generative Adversarial Networks".

adversarialvariationalbayes's Issues

Experiments avb_mnist_8_ac and avb_mnist_32_ac are the same

Hi,

I have looked through the MNIST experiments and I have noticed that experiments 'avb_mnist_8_ac' and 'avb_mnist_32_ac' have exactly the same run.py scripts. Could you double check that the scripts are correct? When I try to modify one of them to have z-dim equal to 8 (instead of 32), I get the following error:

Many thanks!

not TF2.0 compliant

contrib slim doesn't exist for 2.0, and looks to require substantial reworking.

https://www.tensorflow.org/guide/migrate#a_note_on_slim_contriblayers

Why is z^2 added to the adversery?

Specifically here

Bug in interpolation code

Currently, the encoder is used only once for the interpolations, thus resulting in bad reconstructions. This is an error in the github implementation and will be fixed soon.

nan's appearing after 33000 epochs.

I started the run_avae.py script with the default configuration for mnist dataset, however I got nan after 33000 epochs:
ELBO: nan, ELBO (val): nan: 17%|##### | 33703/200000

Is there any specific configuration for the mnist dataset, that I should include when running it?

run_tests.py file is missing

When I run test, I noticed that the run_tests file under avb/validate directory is missing? Therefore,in the test period, the avb/avb/test.py cannot produce results properly.

Conv2DSlowBackpropInput: Size of out_backprop doesn't match computed

Hi,

There is an error when running experiments/avb_mnist_32_ac

InvalidArgumentError (see above for traceback): Conv2DSlowBackpropInput: Size of out_backprop doesn't match computed: actual = 3, computed = 4
	 [[Node: decoder_1/conv_0/conv2d_transpose = Conv2DBackpropInput[T=DT_FLOAT, data_format="NHWC", padding="SAME", strides=[1, 2, 2, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/gpu:0"](decoder_1/stack, decoder/conv_0/weights/read, decoder_1/Reshape)]]
	 [[Node: Mean_8/_4759 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/cpu:0", send_device="/job:localhost/replica:0/task:0/gpu:0", send_device_incarnation=1, tensor_name="edge_6175_Mean_8", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/cpu:0"]()]]