Skip to content

Box Annotator – Drawing Bounding Boxes on Images

This code creates a Box Annotator, which is used to draw bounding boxes on an image based on object detection results.

In simple terms, it helps visualize where objects were detected in an image.

For example, if an AI model detects objects such as:

  • cars
  • people
  • bicycles

the Box Annotator will draw rectangles around those objects.

What is a Bounding Box?

A bounding box is a rectangle drawn around an object detected by an AI model.

Example:

+------------------+
|                  |
|       CAR        |
|                  |
+------------------+

The rectangle shows where the object is located in the image.

Purpose of This Code

The main goal of this code is to:

  1. Take an image
  2. Take object detection results
  3. Draw rectangles around detected objects
  4. Return the annotated image

This helps visualize AI predictions clearly.

Libraries Used

This code uses the following libraries:

  • NumPy → handles numerical data
  • OpenCV (cv2) → draws rectangles on images
  • swatahVision modules → manage detections and annotation tools

Main Class

The main class created in this file is:

BoxAnnotator

This class is responsible for drawing bounding boxes on images.

Configurable Settings

When creating a BoxAnnotator, you can configure a few options.

1. Color

Defines the color used to draw the bounding boxes.

Example:

color = ColorPalette.DEFAULT

Different objects can have different colors.

2. Thickness

Controls how thick the rectangle lines are.

Example:

thickness = 2

A higher value means thicker box borders.

3. Color Lookup Strategy

This decides how colors are assigned to objects.

Available strategies include:

  • INDEX → color based on detection index
  • CLASS → color based on object class
  • TRACK → color based on object tracking ID

Main Function: annotate()

The most important function in this class is:

annotate()

This function draws the bounding boxes.

Inputs

The function takes three inputs:

1. Scene

scene

The image where boxes will be drawn.

The image must be a NumPy array.

2. Detections

detections

This contains information about detected objects.

Each detection includes the coordinates of the bounding box.

Example format:

x1, y1, x2, y2

These represent the corners of the rectangle.

3. Custom Color Lookup (Optional)

custom_color_lookup

This allows users to manually control how colors are assigned.

If not provided, the default color strategy is used.

How the Code Works

The function follows these steps:

Step 1 – Loop Through Detections

The code processes each detected object one by one.

for detection_idx in range(len(detections))

Step 2 – Get Bounding Box Coordinates

For each object, the code extracts the rectangle coordinates.

x1, y1, x2, y2

These represent the corners of the box.

Step 3 – Choose a Color

The color for the bounding box is selected using the color lookup strategy.

This ensures consistent colors across detections.

Step 4 – Draw the Rectangle

The rectangle is drawn using OpenCV:

cv2.rectangle()

The rectangle is placed on the image using the coordinates from the detection.

Step 5 – Return the Annotated Image

After all boxes are drawn, the updated image is returned.

return scene

Example Usage

Example of how this annotator might be used:

import swatahVision as sv

image = ...
detections = sv.Detections(...)

box_annotator = sv.BoxAnnotator()

annotated_frame = box_annotator.annotate(
    scene=image.copy(),
    detections=detections
)

This will return an image with bounding boxes drawn around detected objects.

Workflow

The complete process looks like this:

Input Image
      ↓
Object Detection Model
      ↓
Detections (coordinates of objects)
      ↓
BoxAnnotator draws rectangles
      ↓
Annotated Image Output

Why This Is Useful

Bounding box annotation is important because it helps:

  • visualize AI model predictions
  • debug detection models
  • present results clearly
  • understand where objects are located

Summary

This code creates a tool that:

  • receives detection results
  • draws bounding boxes on images
  • allows customizable colors and thickness
  • returns an annotated image

It is commonly used in computer vision applications such as object detection systems.