Skip to content

Latest commit

 

History

History
53 lines (43 loc) · 1.51 KB

README.md

File metadata and controls

53 lines (43 loc) · 1.51 KB

Font2Font

Yun-Chen Lo, Yi-Ren Chen, Jerry Zhang Jian (NTHU EE)

An enhanced zi2zi project [1] with word-oriented data augmentation, feature combination, and transfer learning.

This repo is based on zi2zi project with incremental boost inspired on recent research and our observation on Chinese Characters.

Folder Description

Font2Font
|	README.md
└──	src/
|	└── zi2zi/					# original zi2zi with data aug options
|	└── zi2zi_hir/					# + combine levels of features
|	└── zi2zi_hir_dis/				# + increase discriminator complexity
|	└── zi2zi_hir_morefilter/			# + alternative ways of complexity boost
|	PastDiscussion/
|	PriorWork/
|	Report/
|	└── Interim/
|	└── Final/

Code Running Example

Train zi2zi 
change directory to src/zi2zi;
$ make all // prepare data, train, and evaluate on demo word

Dependency

  • tensorflow 0.8
  • python 3.x

Major contributions

alt text

Word-oriented data augmentation

alt text

Combine different level feature

alt text

Transfer Learning

alt text

Difficulty-aware adjustment

alt text

Result

alt text

Future Work

Add code or tools to generate compatible Chinese font (.ttf)

Add handwriting dataset and tools for further application exploration