(Face Recognition) Pictures very small / cropped. How to see the full picture?

[aftermathematic]

Is there a way I can see the full picture of a face recognition capture? Pictures are now cropped very small, and provide no context of the person who is at my door.

I'm on version 0.17

[inkeep-ai-assistant[bot]]

\The face recognition images you see are intentionally cropped to just the face region. When you click on a face in the UI, it takes you to the snapshot of the object/general object details, not a full-resolution uncropped version of the face image.(1)

Face recognition in Frigate operates on the stream defined by the detect role in your configuration.(2) The cropped face images are what the face recognition model processes, and this is by design - the embedding models run on a set image size, so larger images would be scaled down anyway.(2)

To get better context around detected faces, you have a couple of options:

1. View the person object snapshot: Clicking on the face should take you to the general object details where you can see the full person detection snapshot, which provides more context than just the cropped face.(1)

2. Increase your detect stream resolution: If the overall image quality is poor, you may want to increase the resolution of your detect stream in your camera's firmware to capture better face details.(1) However, note that motion clarity is more important than extra pixels - over-compression and motion blur are more detrimental to results than resolution.(2)

The cropped face view in the Face Library is specifically designed for the face recognition training and matching process, not for providing full scene context. For context about who is at your door, you'll want to reference the associated person object's snapshot or recording.

📚 Sources:

1. Face recognitions: detections extremely zoomed in and pixelated
2. Face Recognition

---

Have another question? Just tag @inkeep.