Custom Auto-Labeling

Custom Auto-Label is a very powerful tool that helps you automate your data labeling process -- it lets you create a new Auto-Label AI that is fully customized and tailored to your dataset. Even when your data contains very niche, uncommon objects, you can quickly train a customized machine learning model in just a few clicks.

Here's how you can create your own custom AI.

1. Export Manually Labeled Dataset

In order to create a Custom Auto-Label model, you first need to export manually labeled data. Go ahead and export a set of labeled images in your project. Labeling at least a few hundred to a thousand images is highly recommended before you start.

2. Create a Custom Auto-Label

Check that your export is complete under the Export History menu, and then click on the Create a Custom Auto-Label AI button. This will take roughly an hour, depending on the volume of exported data.

Click on the "Create a Custom Auto-Label AI" button after the export is complete
Check that the custom model you've created is ready for use under the Custom Auto-Label menu

3. Inference

You can use the newly created Custom Auto-Label model after configuring it under the Project Configuration page. Please refer to the following guides for configuring and using the Auto-Label function. If you still have a lot of unlabeled images left in the project, we suggest that you apply auto-label to a subset of them (a few hundreds to thousands), so that you can go through this iteration several times and further improve the custom AI's accuracy.

Mapping the Auto-Label AI to each object class in the Project Configuration page
Choose from the list of labels and apply Auto-Label

4. Modify

After applying Custom Auto-Label, audit and fix any errors made by the AI.

5. Export Modified Dataset

Now export a new set of labeled data, including the ones you just created using the steps above.

6. Repeat (Create a new Custom Auto-Label AI)

Again, create another Custom Auto-Label AI in the export history menu. Your exported dataset should now contain more labeled data and should be able to train a Custom AI model that performs better than the one you created above.

You can keep improving the performance of your custom model by repeating the steps 2 through 6 above, and you'll need to spend less time on manual data auditing as the model performance improves.

Currently Custom Auto-Label is available for Enterprise Plan users only. You can also check on your Auto-Label usage under the Billing & Usage menu.

Check on the Auto-Label usage under the Billing & Usage menu