BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Australia/Melbourne
X-LIC-LOCATION:Australia/Melbourne
BEGIN:DAYLIGHT
TZOFFSETFROM:+1000
TZOFFSETTO:+1100
TZNAME:AEDT
DTSTART:19721003T020000
RRULE:FREQ=YEARLY;BYMONTH=4;BYDAY=1SU
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:19721003T020000
TZOFFSETFROM:+1100
TZOFFSETTO:+1000
TZNAME:AEST
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20240214T070247Z
LOCATION:Meeting Room C4.9+C4.10\, Level 4 (Convention Centre)
DTSTART;TZID=Australia/Melbourne:20231214T103500
DTEND;TZID=Australia/Melbourne:20231214T105000
UID:siggraphasia_SIGGRAPH Asia 2023_sess152_papers_452@linklings.com
SUMMARY:IconShop: Text-Guided Vector Icon Synthesis with Autoregressive Tr
 ansformers
DESCRIPTION:Technical Papers\n\nRonghuan Wu, Wanchao Su, Kede Ma, and Jing
  Liao (City University of Hong Kong)\n\nScalable Vector Graphics (SVG) is 
 a popular vector image format that offers good support for interactivity a
 nd animation. Despite its appealing characteristics, creating custom SVG c
 ontent can be challenging for users due to the steep learning curve requir
 ed to understand SVG grammars or get familiar with professional editing so
 ftware. Recent advancements in text-to-image generation have inspired rese
 archers to explore vector graphics synthesis using either image-based meth
 ods (i.e., text → raster image → vector graphics) combining text-to-image 
 generation models with image vectorization, or language-based methods (i.e
 ., text → vector graphics script) through pretrained large language models
 . Nevertheless, these methods suffer from limitations in terms of generati
 on quality, diversity, and flexibility.\nIn this paper, we introduce IconS
 hop, a text-guided vector icon synthesis method using autoregressive trans
 formers. The key to success of our approach is to sequentialize and tokeni
 ze SVG paths (and textual descriptions as guidance) into a uniquely decoda
 ble token sequence. With that, we are able to exploit the sequence learnin
 g power of autoregressive transformers, while enabling both unconditional 
 and text-conditioned icon synthesis. Through standard training to predict 
 the next token on a large-scale vector icon dataset accompanied by textura
 l descriptions, the proposed IconShop consistently exhibits better icon sy
 nthesis capability than existing image-based and language-based methods bo
 th quantitatively (using the FID and CLIP scores) and qualitatively (throu
 gh formal subjective user studies). Meanwhile, we observe a dramatic impro
 vement in generation diversity, which is validated by the objective Unique
 ness and Novelty measures. More importantly, we demonstrate the flexibilit
 y of IconShop with multiple novel icon synthesis tasks, including icon edi
 ting, icon interpolation, icon semantic combination, and icon design auto-
 suggestion.\n\nRegistration Category: Full Access\n\nSession Chair: Haisen
  Zhao (Shandong University)
URL:https://asia.siggraph.org/2023/full-program?id=papers_452&sess=sess152
END:VEVENT
END:VCALENDAR