Object detection and segmentation model
Finds where things are in an image, drawing boxes or exact outlines around each object.
What it is
Detection models locate and label objects with bounding boxes; segmentation models outline them pixel by pixel. They run in cameras, robots, cars and photo apps, often in real time, and promptable models can segment anything you click or describe.
How it works
A backbone network extracts image features, and a head predicts boxes, class labels or masks from them. Single-stage detectors such as YOLO do this in one pass for speed. Promptable segmenters take a point, box or text and return the matching mask.
What it is good for
- Counting and tracking objects in video
- Robotics and self-driving perception
- Background removal and photo editing
- Retail shelf and inventory monitoring
Limits to know
- Struggles with small, hidden or unusual objects
- Needs labeled examples for new classes
- Lighting and camera changes affect accuracy
