Imaging Lunch: Segment-Anything Model (SAM): A Foundation Model for Object Segmentation and Tracking
Event details
| Date | 25.02.2026 |
| Hour | 11:00 › 13:00 |
| Speaker | Pablo Garcia-Amorena, Daniel Sage |
| Location | |
| Category | Conferences - Seminars |
| Event Language | English |
Registration
Abstract
Vision transformers are powering a “ChatGPT moment” for image segmentation: instead of tedious manual outlining, you can steer a model with a few clicks or a short prompt and get usable masks in seconds. In this workshop, we will show how to turn today’s foundation model tools into a practical, repeatable workflow that fits real imaging projects, from quick measurements to scalable annotation.
After a short, intuition-first overview of segmentation using vision transformers and how these models respond to prompts, we will jump into hands-on demos with the latest Meta Segment Anything Models (SAM, including models SAM2 and SAM3) to segment, refine, and propagate masks across image series/video-like data. We will also cover streamlined workflows for dataset creation and user-assisted annotation, plus tips that improve speed and consistency on challenging boundaries and large images.
Finally, we will focus on recognizing common failure modes, validating results, and documenting prompts and settings so segmentations stay reliable and reproducible.
Learning Objectives
About the Imaging Lunches:
Once per month, the EPFL Center for Imaging organises an event dedicated to all PhD students and postdocs working with/in imaging. Discuss the latest advances in imaging. Connect with imaging peers. Learn about popular imaging tools!
Abstract
Vision transformers are powering a “ChatGPT moment” for image segmentation: instead of tedious manual outlining, you can steer a model with a few clicks or a short prompt and get usable masks in seconds. In this workshop, we will show how to turn today’s foundation model tools into a practical, repeatable workflow that fits real imaging projects, from quick measurements to scalable annotation.
After a short, intuition-first overview of segmentation using vision transformers and how these models respond to prompts, we will jump into hands-on demos with the latest Meta Segment Anything Models (SAM, including models SAM2 and SAM3) to segment, refine, and propagate masks across image series/video-like data. We will also cover streamlined workflows for dataset creation and user-assisted annotation, plus tips that improve speed and consistency on challenging boundaries and large images.
Finally, we will focus on recognizing common failure modes, validating results, and documenting prompts and settings so segmentations stay reliable and reproducible.
Learning Objectives
- Use segmentation using foundation models (SAM2/SAM3) to produce and refine masks from clicks/boxes and concept prompts.
- Apply efficient workflows for different production needs: single images, propagation across series/video, and rapid dataset annotation.
- Assess of quality outputs: spot typical failure modes, run quick validation checks and reproducible results.
- Intermediate: comfortable working with scientific images and basic file handling
- Basic Python skills (e.g., running notebooks/scripts, installing packages
- Laptop recommended if you want to follow along hands-o
About the Imaging Lunches:
Once per month, the EPFL Center for Imaging organises an event dedicated to all PhD students and postdocs working with/in imaging. Discuss the latest advances in imaging. Connect with imaging peers. Learn about popular imaging tools!
Links
Practical information
- General public
- Free