Current Result Document :
| 한글제목(Korean Title) |
다양한 물체와 복잡한 배경을 가진 이미지에 적합한 배경 인페인팅 기법 |
| 영문제목(English Title) |
Inpainting Backgrounds on Images with Various Objects and Complicated Backgrounds |
| 저자(Author) |
정혜인
김보은
추연승
김충일
박한무
신사임
Hea In Jeong
Boeun Kim
YeonSeung Choo
Chung-Il Kim
Hanmu Park
Saim Shin
|
| 원문수록처(Citation) |
VOL 45 NO. 01 PP. 1979 ~ 1982 (2022. 06) |
한글내용
(Korean Abstract) |
|
영문내용
(English Abstract) |
Image inpainting is a technique that reconstructs a damaged region within an image. Several deep learning-based approaches have made considerable progress to generate realistic images. However, the conventional approaches tend to experiment on simple datasets, such as CelebA. These datasets only consist of images with similar structures. Hence, some existing methods are not suitable for images with complicated backgrounds. In this paper, we compare the performance of inpainting methods and find a suitable model for images with various objects. Moreover, we experiment on Places2 and Visual Genome to show that training with a dataset which has a large number of objects is better for inpainting real-world images. |
| 키워드(Keyword) |
|
| 파일첨부 |
PDF 다운로드
|
