[AWS Rekognition] Is there need a delay Synchronization / Update issue when index face to collection?

Question

Hi,

So, currently I am implementing a function where it is basically will receive a face, and I will try to find this face in the collection. The flow process will be like this, I will  index (add to collection) the input face to the collection, and get the face_id. Based on the face_id, I will use this face_id to find other faces from the collection (with assumption the collection already store several faces)
The issue for this flow is the generated face_id is not registered in the collection (can't be used), is there any need for delay between index_face and search_face_with_face_id (so I can give some times to AWS Rekognition to sync in the collection data with new face). I directly use the face_id from the index face response and then use it to search face using the face_id. It would be helpful if there is a statement from AWS or documentation that state that we need some delay for the index face to update the collection before the generated face_id can be used to search face.
Thank you

re:Post Agent · Answer

When you index a face using Amazon Rekognition's IndexFaces operation, the face is immediately added to the collection and should be available for searching without any additional delay or synchronization. The FaceId returned by the IndexFaces operation is valid and can be used immediately for subsequent operations like SearchFaces.

However, it's important to note that Amazon Rekognition uses distributed systems, which means that in rare cases, there might be a very brief delay before the indexed face is fully searchable across all of Amazon Rekognition's servers. This delay is typically imperceptible and should not affect most use cases.

If you're experiencing issues where the face ID cannot be found immediately after indexing, it could be due to other factors:

1. Ensure you're using the correct FaceId returned by the IndexFaces operation.
2. Verify that you're searching in the same collection where the face was indexed.
3. Check that you're using the same AWS region for both indexing and searching operations.

If you're still encountering issues, it might be worth implementing a retry mechanism with a short delay (e.g., a few seconds) between attempts. This can help mitigate any potential edge cases related to data propagation in the distributed system.

It's also worth noting that for more robust face recognition, you might want to consider using user vectors instead of individual face vectors. User vectors aggregate multiple face vectors of the same person, which can improve search accuracy significantly compared to using single face vectors.  
**Sources**  
[Amazon Rekognition – frequently asked questions - AWS](https://aws.amazon.com/rekognition/faqs/)  
[Managing face collections, faces, and users - Amazon Rekognition](https://docs.aws.amazon.com/rekognition/latest/dg/managing-face-collections.html)

[AWS Rekognition] Is there need a delay Synchronization / Update issue when index face to collection?

Relevant content