Name: Krisztián Koós
Affiliation: GE Healthcare
Primary research interest: medical image processing, self-supervised learning
Title of the lecture: Interactive Segmentation with Foundation Models
Keywords: segmentation, SAM
Summary: Interactive segmentation is advantageous compared to auto-segmentation methods since the user can affect the result mask with input prompts such as points and bounding boxes. The recent Segment Anything Model (SAM) achieves impressive results in natural image segmentation and can be used in specialized tasks as well which makes it a foundation model. This presentation will cover earlier interactive segmentation approaches, the SAM, and its finetuned medical versions.