Building an image library for custom face detection

Building an Image Library for Custom Face Detection

Curator Version Supported: 3.3 and above

Using AWS Rekognition or Azure AI Face Service, Curator can scan and analyze your media assets to detect faces in images and videos, automatically tagging them with metadata that identifies the specific person (personality) shown.

💡 Note: Custom face recognition is separate from built-in celebrity facial recognition. While AI services have pre-built celebrity libraries, custom face recognition allows users to upload images of any individual to train the AI model. Once trained, Curator works with the AI service to automatically tag relevant metadata to your assets.

For deeper technical details, refer to the official vendor documentation:

• Azure: Azure AI Face Service Documentation
• AWS: AWS Rekognition Collections Documentation

Prerequisites

To enable Curator to recognize and detect custom faces, a System Administrator must first install the appropriate custom face detection service plugin: Azure Custom Face Recognition or AWS Custom Face Rekognition. Custom facial recognition is only possible once an image library is configured using the steps below.

Reference Image Requirements

To train the AI model effectively, the ingested reference assets must meet the following criteria to ensure high-quality analysis results. If these parameters are not met, results may vary and fall short of optimal performance when detecting faces.

• Minimum Size: The face within the image must be at least 200x200 pixels.
• Maximum Resolution: The overall image resolution should not exceed 1920x1080.
• Single Subject: The image must not contain any other faces.
• Full Framing: The image should include the person’s entire head.
• Orientation: The subject must be facing forward.

Creating a Custom Face Image Library

1. Ingest Reference Images

Ingest the high-quality reference image(s) of the individual into Curator.

2. Organize into Collections

• For AWS: Add the image(s) to a collection. We recommend naming the collection after the person (e.g., Jeremy Clarkson).

• For Azure: Add the image(s) to a collection and name it Default (unless a different name is configured within your AZURE-RECOGNITION-UPDATE-CUSTOM-FACE-COLLECTION mediastore under PeopleModeName).
• Note: This single Azure collection can host all your faces and associate them with different personalities.

3. Update Asset Metadata

Select the assets within your collection and update the specific metadata fields based on your cloud provider:

• For AWS: Edit the metadata field FaceCollectionLabel and assign the personality’s name (e.g., Jeremy Clarkson).
• For Azure: Edit the metadata field AzureCustomFacePersonName and assign the personality’s name (e.g., Jeremy Clarkson).

4. Trigger the AI Training Scan

1. Right-click the collection.
2. Select your relevant face recognition service plugin.
3. Choose the option: "Scan collection to update facial recognition library." (Note: Your exact UI plugin options may vary slightly based on your system configuration as shown below)

AWS Plugin Menu

Azure Plugin Menu

5. Verification

Curator will push the FaceCollectionLabel (AWS) or AzureCustomFacePersonName (Azure) metadata to the cloud service.

Once the faces have been confirmed to the library the metadata on the assets within the collection will be updated. FaceCollectionAdded will be updated to True
Once the metadata is set True (Approx 1-3 min), you're all set to begin sending assets for custom face recognition!

Next Steps

To run facial recognition scans on your assets using your newly trained library, please refer to the Face Detection and Celebrity Recognition guide.