AI Paper Atlas

Research map / Image generation / Editing and style transfer

Few-Shot Font Generation With Weakly Supervised Localized Representations

Song Park, Sanghyuk Chun, Junbum Cha, Bado Lee, Hyunjung Shim

TPAMI 2022 · Image generation › Editing and style transfer · approach: Texture and sketch synthesis · 23 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add component-wise style encoder to universal style (ablation)LF-Font · Chinese few-shot font generation, unseen chars (8 refs)style accuracy: 33.6 -> 52.8Improved
use localized instead of universal style representation (ablation)LF-Font · Chinese few-shot font generation, unseen chars (8 refs)style accuracy: 33.6 -> 72.8Improved
add content encoder (ablation)LF-Font · Chinese few-shot font generation, unseen chars (8 refs)harmonic mean accuracy: 21.7 -> 83.2Improved
add component classification loss L_cls (ablation)LF-Font · Chinese few-shot font generation, unseen chars (8 refs)style accuracy: 47.2 -> 72.8Improved
add component classification loss L_cls (ablation)LF-Font · Chinese few-shot font generation, unseen chars (8 refs)content accuracy: 88.6 -> 97.1Improved
add consistency loss L_consist (ablation)LF-Font · Chinese few-shot font generation, unseen chars (8 refs)style accuracy: 69.3 -> 72.8Improved
add consistency loss L_consist (ablation)LF-Font · Chinese few-shot font generation, unseen chars (8 refs)content accuracy: 97.2 -> 97.1No clear effect
increase factor size k from 4 to 8 (ablation)LF-Font · Chinese few-shot font generation, unseen chars (8 refs)harmonic mean accuracy: 82.3 -> 83.2Improved
two-phase instead of end-to-end training (ablation)LF-Font · Chinese few-shot font generation, unseen chars (8 refs)harmonic mean accuracy: 52.2 -> 83.2Improved
use predicted rather than ground-truth character labels at test time (ablation)LF-Font · Chinese few-shot font generation, unseen chars (8 refs)content accuracy: 97.1 -> 94.8Hurt
add persistent memory (DM-Font) in place of content encoder (ablation)LF-Font without content encoder · Chinese few-shot font generation, unseen chars (8 refs)content accuracy: 15.4 -> 53.0Improved
LF-Font vs DG-FontLF-Font · Chinese few-shot font generation, unseen chars (8 refs)harmonic mean accuracy: 63.1 -> 83.2Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy