How mainixkyu/23619007 works
This repository contains a paper review, not an implementation. Its described product is DK-Font: a text-conditioned diffusion model for generating Korean glyphs in a consistent reference style. The documented workflow prepares character and style inputs, combines text, style, and stroke information, and uses a U-Net diffusion model to generate glyphs; training uses MSE and perceptual losses, and results are evaluated with image-quality metrics. Components and links below reflect the README only; no source code was available to verify implementation details.
Main entry points: README.md U-Net diffusion
Components 12
Inputs and preparation
- Korean glyph dataset
README.md - Glyph preprocessing
README.md - Jamo representation
README.md - Reference style image
README.md
Generation model
- Text encoder
README.md - Style encoder
README.md - Character attributes
README.md - U-Net diffusion
README.md - Generated glyphs
README.md
Training and evaluation
- Training losses
README.md - Evaluation metrics
README.md
Other components
- Font designer
Connections 11
- Font designer → Reference style image: provides style
README.md - Korean glyph dataset → Glyph preprocessing: normalizes glyphs
README.md - Glyph preprocessing → Jamo representation: encodes jamo
README.md - Jamo representation → Text encoder: encodes text
README.md - Reference style image → Style encoder: encodes style
README.md - Text encoder → Character attributes: supplies text
README.md - Style encoder → Character attributes: supplies style
README.md - Character attributes → U-Net diffusion: conditions diffusion
README.md - U-Net diffusion → Generated glyphs: generates glyphs
README.md - U-Net diffusion → Training losses: optimizes with
README.md - Generated glyphs → Evaluation metrics: evaluated by
README.md