Just as large language models revolutionized text comprehension and generation, visual foundation models promise to transform the way we approach computer vision tasks. In this roundtable, we share our experience and practical insights from working with recent visual foundation models such as SAM and DINO, exploring how and at which stages they can be integrated into the computer vision pipeline. We will discuss their impact on deep learning workflows, best practices for integration, challenges and limitations, lessons learned from real-world applications, and our expectations for the near future.
Discussion points:
1. Which real-world scenarios have visual foundation models proven most useful for?
2. Do they replace or complement traditional task-specific models?
3. How have they changed data requirements, deployment strategies and efficiency?
4. What are your expectations and concerns for the near future of visual foundation models?
5. How might foundation models create new opportunities in research, industry, or product development?
Laura is a data scientist and computer science graduate from Bar-Ilan University, currently pursuing a PhD in medical imaging at the University of Haifa. Her research focuses on developing deep learning methods for volumetric body composition segmentation in CT scans, with an emphasis on label-efficient training, robustness to noisy labels, and quality control in automated medical image processing pipelines.
Keynote session: Hadas Grossmon Ella
Break
Lightning talks session
Roundtable closing
Talk by Hila Paz
Talk by Dr. Moran Mizrahi
Closing remarks
End
