bbox

A bbox, short for bounding box, is a rectangular region that encloses a set of points or an area in a two-dimensional space. It is a basic, widely used construct in fields such as computer vision, image processing, geographic information systems (GIS), and document analysis. A bounding box typically encodes a location and size of a region of interest.

In computer vision, the axis-aligned bounding box (AABB) is the most common variant. It is defined by minimum and maximum coordinates along each axis, often stored as (xmin, ymin, xmax, ymax) or as (left, top, width, height). Bounding boxes are used to localize objects, crop image regions, and serve as predictions in object detection tasks. Evaluation frequently relies on the Intersection over Union (IoU) metric, which compares predicted boxes to ground-truth boxes. Techniques such as non-maximum suppression (NMS) operate on sets of bounding boxes to reduce duplicates.

In GIS, a bounding box (or extent) describes the geographic territory covering a region, defined by minimum and maximum longitudes and latitudes (minx, miny, maxx, maxy). Bounding boxes are used for spatial queries, map rendering extents, and clipping of raster and vector data. They must be interpreted within a coordinate reference system.

In document analysis and OCR, bounding boxes delineate regions like words, lines, or figures, enabling segmentation, layout understanding, and downstream recognition tasks.

Common representations include (xmin, ymin, xmax, ymax) and (left, top, width, height), with conventions varying by library or domain. Bounding boxes provide a simple, portable abstraction for locating and manipulating regions of interest.