> For the complete documentation index, see [llms.txt](https://docs.unitlab.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.unitlab.ai/documentation/auto-labeling/batch-auto-labeling-with-find-similar.md).

# Batch Auto-Labeling with Find Similar

Use annotated reference images to find and label similar objects across a project, then inspect the results through your workflow.

Use [Find Similar](https://docs.unitlab.ai/documentation/auto-labeling/find-similar) to annotate a batch of images from a few visual examples. Choose one or more reference images, annotate the objects you want to find, assign their classes, and approve the examples. The workflow then uses those saved references to find and label similar objects in other images.

This walkthrough creates **hair** and **shoes** masks from a group photograph, then applies them across a 52-image project. Hair and shoes are examples: use the same process for other objects by supplying representative references for their classes.

## Before you begin

Have your images uploaded or attached to a project, including the images you want to use as references. Prepare the classes in your project's ontology; to follow this example, use **hair** and **shoes** classes with mask annotation. Your workspace role must allow you to manage workflows and access the annotation interface.

## 1. Configure the Find Similar stage

Open **Workflows** in your project and add or select a **Model** stage before **Annotate**. Set **Data type** to **Image**, choose **All project data** under **Batch**, and select **Find similar** under **Model**. You can select a specific queue instead when you want a narrower scope. Leave **Tile scale** on **Auto** for this example, then click **Add reference** in the **References** section.

![](https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/gaXW5KYkmaY7wgy0One9/assets/find-similar-01-border-v2.webp)

*The Model stage is configured for images and is ready to receive reference objects.*

## 2. Choose reference images

In **Choose reference items**, select images containing clear examples of the objects you want to label. The group photograph provides both hair and shoes, so one image can supply examples for both classes. Select one or more images, then click the yellow **Add references** button, which displays your selection count. The reference editor opens in a separate tab.

![](https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/qc1vPdPW1ou9AOhDBePv/assets/find-similar-02-border-v2.webp)

*Select project images that show the appearance and boundaries of your target objects clearly.*

## 3. Check the reference editor

For a new reference, the selected image opens with an empty **Objects** list. Open **Classes** to check that your intended classes are available before you start drawing. For this walkthrough, use the **hair** and **shoes** mask classes. The arrows at the top let you move between selected reference images.

![](https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/lZARx9NDgf9NduW2FmVl/assets/find-similar-03-border-v2.webp)

*The reference editor lets you annotate examples and save them for the workflow.*

## 4. Annotate the first hair example

Zoom in on a clearly visible head and select the **hair** class. Create a mask that follows the hair boundary, keeping the face and background outside it. Check the shape, select the object, and click **Approve** to confirm it as a reference. The object appears in the **Objects** list with its class name and thumbnail.

![](https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/O1X54yDQYgs8HW11EtTG/assets/find-similar-04-border-v2.webp)

*A short hairstyle becomes the first approved hair example.*

## 5. Add another hair example

Keep the **hair** class selected and annotate a second hairstyle with a different appearance. In this example, the longer hairstyle gives Find Similar another visual example of the same class. Refine the mask and approve the second object. Both objects should remain assigned to **hair**, even though their shapes differ.

![](https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/oalCMuwQ1mP3w5gA0gu4/assets/find-similar-05-border-v2.webp)

*Two hair examples represent different lengths and shapes within the same class.*

## 6. Annotate a shoe example

Switch the class selector to **shoes**, then pan or zoom to a clearly visible shoe. Create its mask using the annotation tools; the screenshot shows **SAM 3** selected for assisted annotation. Inspect the proposed boundary, correct it if needed, and approve the object. Check that the new object is listed as **shoes**, alongside the existing hair examples.

![](https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/xLUHAk2sBZqWZkt7i48P/assets/find-similar-06-border-v2.webp)

*The first shoe example adds a second target class to the reference set.*

## 7. Complete the reference set

Add and approve another shoe example, then zoom out to check the set together. The demonstration now contains two hair examples and two shoe examples, all from the same image. You do not need to annotate every matching object in the reference photograph. Each reference image keeps up to **eight approved objects**; approving another after eight replaces the oldest approved example.

![](https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/Qvp1UOfH6uHPiF8rvjbq/assets/find-similar-07-border-v2.webp)

*Four approved objects provide two examples for each target class.*

## 8. Verify approvals and save

Check the **Objects** list before saving: each intended example should have the correct class, a useful thumbnail, and a yellow approval checkmark. If an example is not approved, select it and click **Approve**. Click **Save references** at the top of the editor to make the approved examples available to the workflow. If you selected more reference images, repeat the annotation, approval, and save steps for those images.

<img src="https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/VT6DYTFMFzNvE5BBHjym/assets/find-similar-08-border-v2.webp" alt="" width="188">

*All four example objects are approved and ready to be saved as references.*

## 9. Apply the workflow

Return to the workflow tab and confirm that the saved hair and shoe crops appear under **References**. Check the connections so images pass from **Project** to **Model**, then to **Annotate** for inspection, followed by **Review** and **Complete**. With **All project data** selected, click **Save and Apply** to start Find Similar automatically across all project images routed through the Model stage. You do not need to open or start each image individually. The model uses the saved references to generate annotations, then passes each processed image to **Annotate** for inspection.

![](https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/pNnENQ7yq3SFU7dM8cCY/assets/find-similar-09-border-v2.webp)

*The saved reference crops define the classes and visual examples used by the Model stage.*

## 10. Follow progress in the queue

Open **Queues**, select **Task Queue**, and check the workflow stages. After processing, images appear in **Annotate**, with messages such as **Model added 18 objects**. In this demonstration, the Model stage is empty and all 52 images are in Annotate, ready for inspection. Click **Open** beside an image to check its generated annotations.

![](https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/zaqh3RjMh33YXGGWQKfo/assets/find-similar-10-border-v2.webp)

*The model has handed 52 images to Annotate; annotation checks and review still remain.*

## 11. Inspect the generated masks

Compare the generated annotations with the image, using the **Objects** list to inspect individual hair and shoe masks. Check class assignments, missed objects, duplicate regions, and boundaries that include surrounding pixels. Correct any issues before advancing the image through the workflow. This result shows the saved examples applied to a different group photograph.

![](https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/G42ZPACCfiB13d0Pd2XF/assets/find-similar-11-border-v2.webp)

Example result 1: all hair and shoe annotations shown here were generated automatically by the Find Similar batch run using the saved reference objects.

## 12. Check another image and send to review

Use the top navigation to inspect more images, including different poses, lighting, backgrounds, and object sizes. Check the masks and make any necessary corrections on each image. When an image is ready, click **Send to Review** to move it to the Review stage. Continue through the batch using your project's review process.

![](https://content.gitbook.com/content/GjVLUz4wthGkGlRKM6rM/blobs/N9ZR2Q04pRqT5lupmXOn/assets/find-similar-12-border-v2.webp)

Example result 2: this is another image from the same batch. All hair and shoe annotations shown here were generated automatically from the same saved references.

## Choose useful references

Start with clear examples and accurate boundaries. Add examples that cover meaningful variation, such as different shapes, viewpoints, or sizes, while keeping class assignments consistent. A few carefully selected references are a useful starting point; inspect results to identify where more representative examples may help.

To improve a reference, reopen its crop from the workflow, edit it, and click **Update references**. Use the saved changes for subsequent processing and check the resulting annotations again.

## Related annotation guides

For the annotation tools and review process in each viewer, see [Image Annotation](https://docs.unitlab.ai/documentation/annotations/image-annotation), [Video Annotation](https://docs.unitlab.ai/documentation/annotations/video-annotation), [Medical Annotation](https://docs.unitlab.ai/documentation/annotations/medical-annotation), [Pathology Annotation](https://docs.unitlab.ai/documentation/annotations/pathology-annotation), and [Geospatial Annotation](https://docs.unitlab.ai/documentation/annotations/geospatial-annotation).