Segment Anything Model Quiz

Answers and explanations appear as you take the quiz.

1. What core idea makes SAM a 'foundation model' for segmentation?
   • It treats segmentation as a promptable task that generalizes to unseen objects
   • It only works on faces
   • It requires retraining for every new image
   • It outputs bounding boxes, never masks
2. Roughly how large is the SA-1B dataset used to train SAM?
   • About 1,000 masks
   • Over 1 billion masks across 11 million images
   • About 1 million masks
   • Exactly 50,000 masks
3. Why does SAM split its architecture into a heavy image encoder and a lightweight prompt encoder/decoder?
   • To make training impossible
   • To avoid using any neural networks
   • So one image embedding can be reused for many interactive prompts in real time
   • Because masks are computed on the CPU only
4. How does SAM handle an ambiguous prompt, like a single click that could mean several objects?
   • It refuses to output anything
   • It always picks the largest object
   • It asks the user to type a description
   • It predicts multiple valid masks and ranks them by confidence
5. What type of backbone does SAM use to encode the input image?
   • A Vision Transformer (ViT)
   • A recurrent neural network
   • A decision tree
   • A support vector machine
6. What major capability did SAM 2 add in 2024?
   • Audio transcription
   • Segmenting and tracking objects across video frames
   • Generating text captions
   • 3D printing support
7. What role did the data engine play in the original SAM project?
   • It replaced the image encoder with a text-only model
   • It automatically verified every mask as medically correct
   • It iteratively combined model-assisted annotation and human annotation to build SA-1B
   • It generated labels without human or model assistance
8. How does SAM 2 use memory for video segmentation?
   • It discards all frames after processing each one
   • It predicts only a video-level class label
   • It replaces the prompt with an audio transcript
   • It stores information from prior frames to help segment the object in later frames