FACE FEATURE EXTRACTION SDK

Face detection, alignment and feature extraction.

Detect and align faces, then extract feature vectors inside your own edge application.

Try the online demo Request evaluation Request OEM pricing

From frame to feature

Your application supplies an image or a decoded video frame. The SDK detects faces, aligns face images using the landmarks supported by the model, and returns feature vectors. Video stream access and decoding are part of the surrounding application or a separately scoped integration service.

Image / decoded frame
Detection and alignment
Feature vector output

Designed around your device

The target environment is RK3588 running Linux ARM64. Before supplying an evaluation package, we check the board model, board support package (BSP), RKNN Runtime version, memory and model version. A shared chipset alone does not establish compatibility.

A clear integration boundary

The core interface is a C++ SDK. A local socket or HTTP adapter requires a separately agreed scope. The standard SDK does not include an identity database, 1:1 matching, 1:N search or identity decisions.

Evaluate Face SDK

Conceptual illustration of face detection, alignment and feature-vector output.

Workflow illustration

EXPLORE THE DEMO

Ready to explore the demo?

Open the 88PL experience platform. When you are ready to discuss integration, return here to request an SDK evaluation.

Open demo platform

Online demos and an SDK evaluation are separate steps. Opens in a new tab

Integration details

Specifications and scope

Target platform: RK3588 · Linux ARM64
Core integration: C++ SDK; additional adapters by agreement
Input: Image or decoded video frame; formats confirmed per release
Output: Face detection results and feature vectors; landmark outputs depend on the delivered model and interface
Compatibility review: Board, BSP, RKNN Runtime, memory and model release
Delivery package: Binary libraries, model files, API headers, sample code, documentation and license instructions
Performance: Measured during evaluation with agreed hardware, resolution and concurrency

GOOD TO KNOW

Frequently asked questions

Does local processing mean fully offline operation?

Image inference runs on the target device. Activation and license checks are separate requirements, so local inference alone is not a promise of fully offline operation. If your deployment has no network access, confirm the activation and ongoing license requirements before evaluation.

What is the feature vector dimension?

Feature dimensions, data type, input resolution and preprocessing belong to a particular model release. These are confirmed in the evaluation package rather than assumed across models.

Does it include identity verification and face search?

The standard SDK provides face detection, alignment and feature extraction. It does not return a person’s identity or include 1:1 matching, 1:N search or an identity database. Your application must provide those functions; any additional module or development service needs a separate technical and commercial assessment.

Let’s build your next edge product.

Start with your hardware, your requirements and a focused evaluation.

Request a quote