ICT명품인재양성사업단에서 특강을 실시합니다.
Image Manipulation에 관심 있는 학생들은 누구나 참여가 가능합니다.
1. 일시: 2022. 03. 29(화), 14:00~16:00
2. 제목: Toward a More Practical Image Manipulation / 한국어 강의
3. 강사: 유재준 교수(UNIST)
Abstract
Every recent image manipulation algorithm generally requires a lot of supervision. They basically require many images to train a model, and these images must provide either image-level (i.e. input-output pairs) or set-level (i.e. domain labels) supervision for successful training. However, when collecting data in practice, not to mention the image-level supervision, mere set-level supervision is a severe bottleneck. Even more, there are many important applications (e.g., medical imaging) where there is simply no ground truth available. In this talk, I present my recent works to address this data dependency in various ways; changing model architectures, a new training scheme, data augmentation methods, etc.
Every recent image manipulation algorithm generally requires a lot of supervision. They basically require many images to train a model, and these images must provide either image-level (i.e. input-output pairs) or set-level (i.e. domain labels) supervision for successful training. However, when collecting data in practice, not to mention the image-level supervision, mere set-level supervision is a severe bottleneck. Even more, there are many important applications (e.g., medical imaging) where there is simply no ground truth available. In this talk, I present my recent works to address this data dependency in various ways; changing model architectures, a new training scheme, data augmentation methods, etc.