|
 코드번호 : 36 |
|
| 발표자 : 임화섭 |
|
| 소속 : KIST |
|
| 부서 : |
|
| 직위 : 단장 |
|
| 세션시간 : 14:00~15:50 |
|
| 발표자약력 : |
한국과학기술연구원 인공지능연구단, 2011-현재
영국 Surrey대 방문연구원, 2019-2020
삼성종합기술원 MixedReality Lab, 2007-2011
펜실베니아주립대 전기공학 박사, 2007 |
|
|
| 강연요약 : |
High-quality 3D reconstruction of humans typically requires over 60 synchronized cameras in specialized environments to minimize photometric loss by ensuring the 3D model aligns closely with provided images. The model's quality is adjustable based on capture settings and purpose. Recent generative methods, favored for creating models from text descriptions, employ score distillation sampling loss to precisely align models with their textual inputs. Despite improvements in quality, these methods still face challenges with controllability.
This seminar introduces generative reconstruction techniques that address the challenges of traditional capture setups and improve model controllability. |
|
|
|
