Definition
A bounding box is a rectangular annotation around an object in an image or video. It defines the position and size of the object for training computer vision models.
Purpose
The purpose is to provide labeled examples so AI systems can learn to detect and localize objects in images. Bounding boxes are fundamental in object detection tasks.
Importance
- Simplest and most common annotation type in computer vision.
- Required for training models like YOLO or Faster R-CNN.
- Limited in precision for irregularly shaped objects.
- Basis for advanced annotations like polygons or masks.
How It Works
- Define object categories for detection.
- Draw rectangles around objects in images.
- Record coordinates as annotation data.
- Validate with quality checks.
- Use annotated data to train object detection models.
Examples (Real World)
- COCO dataset: bounding box annotations for common objects.
- Tesla Autopilot: bounding boxes for vehicles and pedestrians.
- Amazon Go stores: bounding boxes used in computer vision checkout systems.