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:20260114T163648Z
LOCATION:Meeting Room C4.9+C4.10\, Level 4 (Convention Centre)
DTSTART;TZID=Australia/Melbourne:20231214T145000
DTEND;TZID=Australia/Melbourne:20231214T150000
UID:siggraphasia_SIGGRAPH Asia 2023_sess132_papers_430@linklings.com
SUMMARY:Domain-Agnostic Tuning-Encoder for Fast Personalization of Text-To
 -Image Models
DESCRIPTION:Moab Arar (Tel-Aviv University); Rinon Gal (Tel Aviv Universit
 y, NVIDIA Research); Yuval Atzmon (NVIDIA Research); Gal Chechik (NVIDIA R
 esearch, Bar-Ilan University); Daniel Cohen-Or (Tel Aviv University); Arie
 l Shamir (Reichman University (IDC)); and Amit H. Bermano (Tel Aviv Univer
 sity)\n\nText-to-image (T2I) personalization allows users to guide the cre
 ative image generation process by combining their own visual concepts in n
 atural language prompts. \nRecently, encoder-based techniques have emerged
  as a new effective approach for T2I personalization, reducing the need fo
 r multiple images and long training times.\nHowever, most existing encoder
 s are limited to a single-class domain, which hinders their ability to han
 dle diverse concepts. In this work, we propose a domain-agnostic method th
 at does not require any specialized dataset or prior information about the
  personalized concepts. We introduce a novel contrastive-based regularizat
 ion technique to maintain high fidelity to the target concept characterist
 ics while keeping the predicted embeddings close to editable regions of th
 e latent space, by pushing the predicted tokens toward their nearest exist
 ing CLIP tokens. Our experimental results demonstrate the effectiveness of
  our approach and show how the learned tokens are more semantic than token
 s predicted by unregularized models. This leads to a better representation
  that achieves state-of-the-art performance while being more flexible than
  previous methods.\n\nRegistration Category: Full Access\n\nSession Chair:
  Jun-Yan Zhu (Carnegie Mellon University)\n\n
URL:https://asia.siggraph.org/2023/full-program?id=papers_430&sess=sess132
END:VEVENT
END:VCALENDAR