LabelMe is an actively developed open source graphical image annotation tool inspired by the app of the same name released in 2012 by MIT CSAIL.
It is capable of annotating images for object detection, segmentation, and classification (along with polygon, circle, line, and point annotations). It also supports annotating videos.
The app is cross-platform and runs on Ubuntu, macOS, and Windows using Qt4 (or Qt5) and Python (2 or 3). A docker container is also available.
Also check out our LabelMe Tutorial.
Below, learn the structure of LabelMe JSON.
{
"version": "4.5.6",
"flags": {},
"shapes": [
{
"label": "black-bishop",
"points": [
[
250.5,
44.0
],
[
285.0,
112.5
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "white-knight",
"points": [
[
128.5,
44.0
],
[
168.5,
110.5
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "white-king",
"points": [
[
110.5,
153.5
],
[
160.0,
253.5
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "white-bishop",
"points": [
[
91.0,
295.0
],
[
136.5,
366.0
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "white-pawn",
"points": [
[
242.0,
253.5
],
[
277.5,
306.5
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "white-queen",
"points": [
[
317.0,
33.0
],
[
357.0,
117.0
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "white-bishop",
"points": [
[
367.5,
1.5
],
[
401.0,
69.5
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "black-rook",
"points": [
[
484.5,
56.5
],
[
522.5,
113.0
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "black-king",
"points": [
[
557.0,
108.0
],
[
614.5,
206.0
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "black-pawn",
"points": [
[
384.0,
153.0
],
[
412.5,
204.5
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "black-bishop",
"points": [
[
388.5,
242.0
],
[
428.0,
310.5
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "black-pawn",
"points": [
[
517.5,
249.5
],
[
553.0,
303.5
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
},
{
"label": "black-knight",
"points": [
[
466.5,
356.5
],
[
516.0,
430.5
]
],
"group_id": null,
"shape_type": "rectangle",
"flags": {}
}
],
"imagePath": "0af.jpg",
"imageData": "/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAgGBgcGBQgHBwcJCQgKDBQNDAsLDBkSEw8UHRofHh0aHBwgJC4nICIsIxwcKDcpLDAxNDQ0Hyc5PTgyPC4zNDL/2wBDAQkJCQwLDBg...",
"imageHeight": 512,
"imageWidth": 765
}With Roboflow supervision, an open source Python package with utilities for completing computer vision tasks, you can merge and split detections in LabelMe JSON. Read our dedicated guides to learn how to merge and split LabelMe JSON detections.
Below, see model architectures that require data in the LabelMe JSON format when training a new model.
On each page below, you can find links to our guides that show how to plot predictions from the model, and complete other common tasks like detecting small objects with the model.