Learning Eastern Asian language fonts with conditional adversarial networks. This is an extension of the recent work "Image-to-image translation with conditional adversarial networks" by Isola P. et al.

Currently, the network is only capable of performing one-to-one font style transfer. We may extend it to support one-to-many style transfer in the future.

Notably, the network has never seen any Korean characters before the inferring stage. The experiment shows that our network is capable of performing style transfer on previously unseen characters.

• Linux or OSX
• NVIDIA GPU + CUDA CuDNN

• Download the fonts (.ttf or .otf)
• Generating corresponding images for the fonts, call:

python script/generate.py --src_font=xxx.ttf --dst_font=xxx.otf --charset=CN --shuffle=1 --filter=1 --sample_count=3000 --sample_dir=.

DATA_ROOT=/path/to/data/ name=expt_name which_direction=AtoB th train.lua

Switch AtoB to BtoA to train translation in opposite direction. See opt in train.lua for additional training options.

DATA_ROOT=/path/to/data/ name=expt_name which_direction=AtoB phase=val th test.lua

See opt in test.lua for additional testing options.