Variational Grid Setting Network
This work addresses a domain-specific problem for font design and typography, offering an incremental improvement by adapting existing methods to generate high-resolution Chinese characters with limited data.
The authors tackled the problem of automatically generating missing Chinese characters in a font set by proposing the Variational Grid Setting Network, a neural network architecture based on a tweaked variational autoencoder, which can generate large 256x256 pixel characters and achieve satisfactory results with very few training samples.
We propose a new neural network architecture for automatic generation of missing characters in a Chinese font set. We call the neural network architecture the Variational Grid Setting Network which is based on the variational autoencoder (VAE) with some tweaks. The neural network model is able to generate missing characters relatively large in size ($256 \times 256$ pixels). Moreover, we show that one can use very few samples for training data set, and get a satisfied result.