SAFR™ for Camera-Equipped Devices
Integrate world-class facial detection and recognition into products and solutions at the edge or entirely offline. Request an invitation to try SAFR SDKs and Web APIs free for 30 days.
The SAFR SDK
Access the full capabilities of the SAFR platform through RESTful APIs on all major platforms and operating systems
The SAFR Embedded SDK
Add high-performance detection and recognition to edge devices that have resource constraints and lack connectivity
Five Reasons to Develop with the SAFR SDKs:
- World-class accuracy and performance for edge and IoT devices
- A fraction of the cost of cloud services and price certainty
- Data is private and secure to protect your customers
- Easily incorporate into your stack
- Innovation from an established US-based company
The SAFR SDK
Architected to address the detection and recognition needs of developers building new solutions or extending existing products for a wide range of use cases.
Detection and tracking in live video streams
Recognition and DB matching
Characterization (age, gender, sentiment)
Identity repository and management
SAFR Actions execution
Video feed management
99.87% LFW, 0.0334 FNMR
<100ms per face
~15-60ms, @1024p and 15FPS
The SAFR Embedded SDK
Optimized for embedded systems and camera-equipped devices with resource and network constraints providing an unprecedented blend of accuracy, speed, and efficient size.
Face signature store (+/- 10,000 signatures)
CLI access to image and DB parameters
99.87% LFW, 0.0334 FNMR
900ms @720p on a single core, 1.8MHz ARM processor
300ms @720p on a single core, 1.8MHz ARM processor
Frequently Asked Questions
How do I get the SDKs?
You can get access to the SAFR SDKs by requesting an invitation. Once approved, your 30-day free trial begins.
Where do I find documentation for the SDKs?
All SDK-related documentation, including quick-start guides, links to sample apps, and platform specs, can be found at the developer documentation site. You can also find our Web API-related documentation and, once you’ve signed up for a developer account, you can follow the links to try them out.
Which SDKs does SAFR offer?
The SAFR SDK allows you to run face detection and tracking on-device. For recognition, the SAFR SDK performs identity management and provides callbacks for recognition events. The SAFR SDK is deployable on-premise, in your cloud, or in the SAFR cloud.
The SAFR Embedded SDK allows you to run face detection, tracking, and recognition entirely offline and on-device. The SAFR Embedded SDK is most suitable for devices with low bandwidth, limited power, and memory constraints.
Which platforms are the SDKs available on?
Which formats does the SAFR SDK support as input?
SAFR SDK is compatible with MP4, H.264, and H.265 video file formats, while SAFR Embedded SDK does not support video. SAFR SDK and SAFR Embedded SDK both require you write your own code to use image files like JPG and PNG.
Does the SAFR SDK work with network or USB cameras?
SAFR SDK is compatible with any IP camera or USB camera.
What can I do with the SDKs?
The SAFR SDK offers the following features:
- Face Detection: Find faces in still images and live/archived video streams.
- Face Tracking: Track faces and certain objects in live video streams.
- Face Recognition: Look for known faces — matched from an existing database — in still images and live/archived video streams.
- Event generation and reporting: Record events, including time-stamps, when a specified person or a group of specified people appear on a video stream. Events can be programmed to trigger real-time alerts.
- Person Characterization: Predict gender, sentiment, and approximate age for each detected face.
- Identity repository: Add, delete, or retrieve stored identities — including associated metadata.
- SAFR Actions execution: Configure alerts and enable powerful custom actions and alarms based on recognition events — unlock a door, turn on lights, sound an alarm, deny a person entry to a secure access point, log demographic information, generate reports, notify police, deploy security guards, initiate a building lockdown, and more.
- Camera discovery and connection: Look for available cameras on a network and easily setup USB, IP, or ONVIF cameras.
- Customizable face image quality metrics: Setup thresholds to determine the minimum quality bar of incoming images or video feeds before face detection/recognition is performed.
The SAFR Embedded SDK offers the following features:
- Completely offline Face Detection: Find faces in still images, without having to connect to the internet.
- Completely offline Face Recognition : Look for known faces — matched from an existing database — in still images, without having to connect to the internet.
- Face signature store (+/- 10,000 signatures): This store is local, on-device.
- CLI access to image and DB parameters: Learn and recognize persons in an image and manage person records and metadata in the database.
What else do I need beside the SAFR SDK or SAFR Embedded SDK to perform facial recognition?
For both SDKs, you will need any IP or USB camera. If using the SAFR SDK, in addition to the hardware you’re running the SAFR SDK on, you will also need separate hardware to run the SAFR Server. The SAFR Embedded SDK can work in environments with limited resources or network constraints, and does not require any additional software.
Do you have an offline SDK?
Yes! The SAFR Embedded SDK runs entirely offline, performing both detection and recognition without an internet connection.
How much do the SDKs cost?
You are welcome to try both SDKs for 30 days free of charge. Each SDK offers monthly tiered pricing based on the number of live feeds. Detailed pricing is available on request. Please contact us to learn more.
What kind of speed and accuracy can I expect with the SAFR SDKs?
SAFR matches against millions of faces in under one second with 99.86% accuracy. In the July 2019 National Institute of Standards and Technology (NIST) results, SAFR tested as both the fastest and most compact among algorithms for wild images with a less than 0.0335 False Non-Match Rate. Compared to the other high-accuracy algorithms, SAFR is 2x faster than average, and also has one of the lowest rates of bias with respect to skin tone and gender. SAFR performs consistently across a range of skin tones due to its massive and highly diverse global training set.
How secure is my data?
All facial signatures and facial images use AES-256 encryption, in transit or at rest, and no data is passed over the internet if run on prem. Privacy is a top priority, from our full suite of administrative tools, including easy opt-in/opt-out, to thoughtfully adhering to our Implementation Best Practices.
Do the SAFR SDKs offer people counting?
People counting is not offered through the SAFR SDK at this time. However, it has been added to our applications and will follow in our SDKs in the near future.
Do I need to run the SAFR Server?
If you’re using the SAFR SDK, it depends whether you’re using the SAFR Cloud or the on-prem SAFR Server to perform recognition. If you’re using the SAFR Embedded SDK, no additional software is required.
How does SAFR compare with other facial recognition SDKs?
- Significantly lower total cost of ownership
- Can be easily integrated into camera-equipped devices
- Flexible deployments including online and totally offline
- The Embedded SDK has a small footprint, making it suitable for memory or power constrained devices
- All data remains on the device to ensure privacy
Is SAFR accessible via REST APIs?
Yes, the SAFR Computer Vision REST API enables face detection, face recognition, and management of recognized persons and faces. Additional interfaces provide object recognition, configuration, and user management. Request and invitation for a developer account to test SAFR REST APIs.
Let’s Get Something Started
Request an invitation to try SAFR SDKs and Web APIs free for 30 days
Contact us for more information about SAFR