| 1. 제 목 : Style-Controlled Synthesis of Clothing Segments for Fashion Image Manipulation
2. 발 표 자 : 김보경 연구원 (KAIST 전기및전자공학부)
3. 일 시 : 2019년 08월 07일(수) 16:00 ~ 18:00
4. 장 소 : 경북대학교 IT-1호관 318호
5. 초청교수 : 이민호 교수
6. 주관 : 경북대학교 인공지능연구원, KNU-LG전자 융합연구센터
7. 강사약력 :
Bo-Kyeong Kim is a Ph.D. candidate in the School of Electrical Engineering at Korea Advanced Institute of Science and Technology (KAIST).
She received the B.S. degree in Bio and Brain Engineering from KAIST in 2012. Her research interests include machine learning and image processing.
8. 내용요약 :
Fashion is part of our everyday lives, and people are interested in knowing how they would look in different clothes. In this talk, I present an approach for digitally altering people's outfits in images. Given images of a person and a desired clothing style, this method generates a new clothing item image. The new item displays the color and pattern of the desired style while geometrically mimicking the person's original item.
Through superimposition, the altered image is made to look as if the person is wearing
the new item.
Unlike recent works with full-image synthesis, this method relies on segment synthesis, yielding benefits in virtual try-on. In the synthesis process, two factors characterizing clothing segments (i.e., geometry and style) are disentangled via preprocessing and combined using a neural network. Experimental results show
(1) the generation of high-quality clothing segments,
(2) the applicability of this method for virtual clothing try-on, and
(3) visual differences to existing neural style transfer in changing clothing styles.
◀ 문의처 : IT대학 전자공학부 인공두뇌연구실 장주희 ☎ 940-8616 ▶
|
