Bounding Box

Overview

Bounding boxes are the simplest form of annotation you can place on an image, created by marking the two corners of a box. Bounding boxes in labelbox are designed to be easy to create and manipulate.

Configuring Bounding Box

Interactive Configuration

Programmatic Configuration

"required"

indicates whether or not this tool is required to submit the label

"name"

The name of the tool

"color"

a color name or hex code for the label display

"tool"

what type of tool you’re using, in this case "rectangle"

"Classifications"

any nested classification on the bounding box

"tools": [
{
"required": false,
"name": "Car",
"color": "#039BE5",
"tool": "rectangle",
"classifications": []
}
]

Nested Classifications

Any existing classification tools can be nested within the tool to capture metadata about the label. Below is an example nesting single choice question (radio) within the bounding box tool.

Interactive Configuration

Programmatic Configuration

"tools": [
{
"required": false,
"name": "Car",
"color": "#039BE5",
"tool": "rectangle",
"classifications": [
{
"name": "occluded?",
"instructions": "Occluded?",
"type": "radio",
"options": [
{
"value": "yes",
"label": "Yes"
},
{
"value": "no",
"label": "No"
}
],
"required": false
}
]
}
]

Label Format

The bounding box can be imported as an existing label, an model prediction or exported in the default JSON format as shown below. The coordinates are in the clockwise direction and the last coordinate isn’t the same as the first.

"Car": [{
"geometry": [{
"x": 453,
"y": 334
}, {
"x": 453,
"y": 796
}, {
"x": 1234,
"y": 796
}, {
"x": 1234,
"y": 334
}]
}]


How did we do?