The VAE has a modular design. The encoder, decoder and VAE are 3 models that share weights. After training the VAE model, the encoder can be used to generate latent vectors. The decoder can be used to generate font images by sampling the latent vector from a Gaussian distribution with mean=0 and std=1.
| encoder | decoder |
|---|---|
 |
 |
dataset size = train 5000 & validation 1000 per each class
width, height = 112, 112
font size = 25
used characters = "AaBbCcDdEeFfGgHhIiJjKkLlMmNnOoPpQqRrSsTtUuVvWwXxYyZz"
images are generated with font2png.py
fonts are from Google Fonts
| idx | font name | sample images |
|---|---|---|
| 0 | EBGaramond |      |
| 1 | PT_Serif |      |
| 2 | NotoSans |      |
| 3 | Roboto |      |
| 4 | Righteous |      |
| 5 | Bangers |      |
| 6 | Pacifico |      |
| 7 | DancingScript |      |
| 8 | Inconsolata |      |
| 9 | VT323 |      |
| epoch | 20 | 50 | 200 |
|---|---|---|---|
| log |  |
 |
 |
| latent space |  |
 |
 |
1 frame = 1 batch
1 epoch = 196 batchs
| epoch=1 | epoch=2 | epoch=3 |
|---|---|---|
 |
 |
 |
| epoch=4 | epoch=5 | epoch=6 |
 |
 |
 |
[1] Kingma, Diederik P., and Max Welling. "Auto-encoding variational bayes."
[2] Keras Blog: Building Autoencoders in Keras
[3] Keras example: VAE
[4] Erik Bernhardsson. "Analyzing 50k fonts using deep neural networks"
[5] TJ TORRES. "A Fontastic Voyage: Generative Fonts with Adversarial Networks"
[6] Julien Despois. "Latent space visualization — Deep Learning bits #2"