Detect and track
One person, one track.
Faces are found and followed through motion, occlusion and changing viewpoints, so each person is handled as a sequence.
Synthetic face replacement for video that needs to stay useful. Built around identity suppression, temporal consistency and retained analytical signals.
In developmentRequest access
SourceResultTest 04 of 10In development
- Kept
- Head pose, gaze direction, lighting, clothing
- Changed
- Facial identity
A blurred face hides who someone is, and also where they are looking, how their head is turned and how they are moving. Synthetic replacement aims to remove only the identity.
Blur
SyntheticThree stages, each with its own evaluation. The research challenge is balancing them.
Replace consistently
A new identity that holds.
A synthetic face is generated and kept stable for the whole track, so there is no frame-to-frame flicker between identities.
Evaluate the trade-off
Measure what was lost.
Recognition suppression is measured alongside pose, gaze and downstream detection on the same footage.
Facilities and public spaces, mobility fleets, transit operators, event production and video archives.
Planned interfaces
# Planned live ingestion aynvia stream \ --engine faceanon \ --in rtsp://10.0.4.12/cam-07 \ --out rtsp://0.0.0.0:8554/cam-07-anon
These are the numbers we are working toward. None of them is a published result or a customer guarantee.
Object-detection recall above 98% on transformed footage, compared with the original.
- Validation needed
- A defined dataset, detector and operating conditions.
Identity-match confidence below 4% against ArcFace and CosFace-based recognition.
- Validation needed
- A calibrated score, recognition thresholds and a published protocol.
Head pose (roll, pitch, yaw) and gaze direction preserved.
- Validation needed
- Measured deviation between source and transformed footage.
Each synthetic identity stays stable for its whole track.
- Validation needed
- Evaluation across motion, occlusion, re-entry and lighting changes.
Can we use FaceAnon today?
Not yet. FaceAnon is in development. Requesting access helps us understand your evaluation needs; it does not give immediate product access.
Does synthetic replacement guarantee anonymity?
No. Replacing a face is one part of a privacy workflow. Clothing, gait, location, metadata and context can still identify someone, and replacement alone does not establish legal compliance.
Why not just blur faces?
Blur and pixelation remove the signals that analysis depends on, such as where someone is looking and how their head is turned. Synthetic replacement aims to keep those signals while removing the identity.
How will it be deployed?
The plan is live RTSP ingestion for cameras and fleets, a batch REST API for archives, and a Python SDK for data pipelines. All three are planned interfaces.
Tell us what you need to protect, extract or inspect. We'll tell you honestly whether our engines can help yet.