InfoGAN for temporal series generation

Introduction

Use infoGAN to generate multi-dimentional temporal sequences.

In MNIST, continuous noise can represent character width and rotation angle. But in here, their meanings are not quite intuitive.

In MNIST, modifying continuous noise would not change the classtype of generated samples. But in here, it happens.

See codes for MNIST at https://github.com/SongDark/GAN_collections.

Datasets

Name	Link	Class	Dimension	Train Size	Test Size	Truncated
CharacterTrajectories	Download	20	3	1422	1436	182

Generate npz file

Unzip CharacterTrajectories.zip at data/CharacterTrajectories, then run dataprocess.py.

    python dataprocess.py

Generated Images

Random noise, Random discrete code, Fixed continuous code

Epoch 0	Epoch 200	Epoch 500

Specified condition, Random noise

Epoch 0	Epoch 200	Epoch 500

Fixed noise

It seems that the generated sequences are not corresponding to their one-hot labels.

label=14(s)	label=16(v)	label=16(y)

Reference

https://github.com/hwalsuklee/tensorflow-generative-model-collections/blob/master/infoGAN.py

https://github.com/buriburisuri/timeseries_gan/blob/master/train.py