For object detection (i.e., bounding boxes), labelme
uses JSON files, storing the objects as list under the shapes
property.
Each shape has a label
and a points
property. The latter is a list of
[xmin, ymin]
and [xmax, ymax]
. The shape_type
is rectangle
.
{
"version": "4.0.0",
"flags": {},
"shapes": [
{
"label": "person",
"points": [
[
191.0,
107.36900369003689
],
[
313.0,
329.36900369003695
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "person",
"points": [
[
365.0,
83.0
],
[
500.0,
333.0
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
}
],
"imagePath": "2011_000003.jpg",
"imageData": null,
"imageHeight": 338,
"imageWidth": 500
}
https://github.com/wkentaro/labelme/tree/main/examples/bbox_detection