Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Explore a 14-minute technical video breakdown of Meta's Segment Anything Model 2 (SAM2), which extends the revolutionary SAM technology from image to video segmentation. Learn about the challenges of video segmentation, understand the model's architecture including the image encoder, memory encoder, memory bank, and memory attention mechanisms. Discover how the data engine generates the largest video dataset to date (SA-V dataset), and examine the experimental results that demonstrate SAM2's capabilities. Delivered by a machine learning researcher with 15 years of software engineering experience and a Master's in Computer Vision and Robotics, dive deep into the technical components of promptable visual segmentation and the end-to-end architecture that makes video object segmentation possible.
Syllabus
- Intro
- Challenges with video segmentation
- Overview of SAM2
- Promptable Visual Segmentation
- SAM2 Model
- End to end architecture
- Image Encoder
- Memory Encoder
- Memory Bank
- Memory Attention
- Training
- Data Engine
- Segment Anything Video SA-V dataset
- Experiments
Taught by
AI Bites