Home Computer Vision LabelMe for Picture and Video Annotation

LabelMe for Picture and Video Annotation

LabelMe for Picture and Video Annotation


What’s LabelMe?

LabelMe is an open supply picture annotation instrument, impressed by MIT’s net annotation instrument of the identical title. LabelMe means that you can create annotations for object detection, classification, and segmentation laptop imaginative and prescient datasets. You may draw annotations utilizing polygon, rectangle, circle, line, and factors.

On this information, we’re going to present methods to use LabelMe. We are going to stroll by:

  1. The best way to set up LabelMe
  2. The best way to add photos
  3. The best way to create an annotation in LabelMe
  4. The best way to export information
  5. The best way to convert LabelMe annotations to completely different codecs

With out additional ado, let’s get began!

LabelMe Options

LabelMe supplies a graphic interface by which you’ll annotate photos. With LabelMe, you possibly can annotate photos for:

  1. Object detection, utilizing bounding packing containers and polygons;
  2. Occasion segmentation;
  3. Semantic segmentation, and;
  4. Classification.

You may as well annotate video information in LabelMe.

The best way to Set up LabelMe

LabelMe gives each working system-specific set up strategies in addition to a Python set up technique. The Python set up technique can be utilized throughout Home windows (by way of WSL), Linux, and macOS, so we are going to use that technique on this information.

LabelMe beforehand provided a hosted net interface for annotating photos, however the net instrument is now not open to new customers.

To put in LabelMe utilizing Python, run the next command:

pip set up labelme

When this command has completed operating, you can begin LabelMe utilizing the next command:


LabelMe will begin and pop up in a brand new window:

With LabelMe put in, we will begin annotating photos.

Add Photographs to LabelMe

To annotate photos in LabelMe, click on “Open Dir” and select a folder of photos you wish to open. You may as well click on “Open” to pick particular person photos.

The folder of photos (or particular person picture) you’ve gotten opened in LabelMe will load. On this instance, we are going to load a folder of photos. After selecting a folder, a picture seems that we will annotate. The sidebar exhibits a “File Record” field in which you’ll navigate between photos within the folder you’ve gotten chosen.

When you want photos to your venture, take a look at Roboflow Universe, a neighborhood for sharing laptop imaginative and prescient datasets. There are over 200,000 public datasets on Universe that you should utilize to assist along with your initiatives.

We now have photos prepared for annotation in LabelMe.

The best way to Create an Annotation in LabelMe

To create an annotation, click on the “Create Polygons” button within the software sidebar, then click on “Edit” within the software command tray to decide on what sort of polygon you wish to create. Right here is how the command tray seems on macOS (the tray will seem in a different way relying on the working system you might be utilizing):

Right here, we will click on “Create Rectangle” to pick the rectangle instrument to be used in drawing a bounding field. After deciding on the rectangle instrument, you possibly can click on anyplace on the loaded picture to begin drawing a bounding field:

To attract a bounding field, click on on a degree, then drag your cursor to attract the field. When you’re completed, click on your mouse or trackpad to set the field. Then, you’ll be requested to pick a category for the article. You may add new lessons as you go, and reuse present lessons you’ve gotten set.

If you’re new to labeling information, take a look at our The best way to Label Photographs for Pc Imaginative and prescient information. The information is stuffed with finest practices that may assist information you to create high-quality annotations to be used in coaching laptop imaginative and prescient fashions.

The record of labels utilized in your venture will probably be saved within the “Label Record” column in the appropriate panel in LabelMe.

To save lots of an annotation, press Command + S (macOS) or Management + S. It will save a JSON file along with your annotation information in the identical folder because the picture on which you might be working is saved. For instance, if you’re working in a folder referred to as “practice”, the annotation will probably be saved in that folder.

The best way to Convert LabelMe Knowledge

LabelMe saves annotations in a proprietary format. Annotated picture information is saved in its personal file. To make use of your annotations in coaching a pc imaginative and prescient mannequin, you’ll probably should convert the annotations to a different format. With assist from Roboflow, you are able to do this in a couple of clicks without cost.

Create a free Roboflow account, then click on “Create New Venture” on the Roboflow dashboard. Set a reputation to your venture and select the kind of venture on which you might be working (object detection, classification, segmentation). Then, click on “Create Public Venture”.

A field will seem in which you’ll drag and drop your photos and annotations from LabelMe. If you drag your information into the field, a pop up will ask if you wish to use the embedded photos within the LabelMe annotations. Click on “Use the Embedded Photographs”:

Upon getting uploaded your information, click on Generate within the sidebar to create a brand new venture model. At this step, you possibly can select to use any preprocessing steps and augmentations to your dataset. When you’re able to create your dataset, click on “Generate”. It will create a frozen-in-time model of your information which you can export.

When your dataset has been generated, you possibly can select to export the info in numerous completely different codecs (i.e. COCO, Pascal VOC):

LabelMe Alternate options

Inside LabelMe, you possibly can annotate polygons with a easy level and click on. However, LabelMe will not be the one instrument accessible for this job. There are many different annotation instruments you should utilize to your laptop imaginative and prescient venture.

See how LabelMe stacks up in opposition to different annotation instruments by trying out our different weblog posts on various instruments:


LabelMe is a free, open-source annotation instrument that you should utilize for labeling laptop imaginative and prescient datasets. On this information, we’ve demonstrated methods to use LabelMe. We walked by putting in LabelMe, making a label, and how one can convert LabelMe information to completely different codecs.

Incessantly Requested Questions

Is LabelMe free to make use of?

Sure. LabelMe is an open supply instrument and is free to make use of.

Does LabelMe have a hosted net interface?

The hosted LabelMe net interface is now not formally supported, however you possibly can nonetheless use the desktop software.

James Gallagher.” Roboflow Weblog, Jul 20, 2023. https://weblog.roboflow.com/labelme/



Please enter your comment!
Please enter your name here