Deprecated featureThis feature has been deprecated. Please see below for additional details.
Prerequisites
- Make sure your project belongs to an organization. See transfer ownership for more info.
- Make sure your project is configured as an object detection project. You can change the labeling method in your project’s dashboard. See Dashboard for more info.
- Add some images to your project, either by collecting data or by uploading existing datasets. See Data acquisition for more info.
- You now should be able to see the Auto-labeler tab in your Data acquisition view:

Data acquisition with auto-labeler available when using an enterprise project
Object detection auto-labeler settings
Which items to include:- All data items present in your dataset
- Data items in the labeling queue
- Data items without a given class

Auto-labeler settings
Note that this process is slow (a few seconds per image, even on GPUs). However, we apply a strong cache on the results, so once you have ran the auto-labeler once, your iterations will be must faster. This will allow you to change the settings with less friction.
Label clusters
Once the process is finished, you will be redirected to a new page to associate a label with a cluster:
Add a label to a cluster
Example
Each project is different, to write this documentation page, we have collected images containing several dice. This dataset can be used in several ways - you can either label the dice only, the dice color or the dice figures. You can find the dataset, with the dice labeled per color in this public project. To adjust the granularity, you can use the Sim threshold parameter.1. Group all the dice together:
Here we have been setting the Sim threshold to0.915

Auto-labeler clusters
2. Group the dice by color:
Here we have been setting the Sim threshold to0.945

Auto-labeler clusters
3. Group the dice by color and by figure:
Here we have been setting the Sim threshold to0.98

Auto-labeler clusters

Model trained with FOMO