3D Face SDK

1. High precision and high robustness

A 3D face reconstruction algorithm based on deep neural network optimization can accurately restore over 1000 3D keypoints of the face, with a reconstruction error of ≤ 1mm. It supports stable detection in multi angle (± 90 ° side face), multi lighting (strong/weak/backlight), and multi occlusion (mask/glasses) scenes;

Adapt to the hardware differences of Android devices, automatically adapt to RGB cameras+depth sensors (such as TOF, structured light) or pure RGB monocular vision solutions, and achieve high-precision 3D face reconstruction through monocular video streams on devices without depth hardware.

2. Lightweight and high-performance

Lightweight algorithm tailoring for Android mobile devices, with a core model size of ≤ 50MB and memory usage of ≤ 200MB, meeting the storage and runtime resource limitations of mobile devices;

Real time processing frame rate ≥ 30fps (mid-range Android models), 3D face modeling time ≤ 300ms, no visible graphics lag, meeting the requirements of real-time interactive experience on mobile devices;

Support hardware acceleration for Android systems (such as NNAPI, OpenCL), fully utilize the computing power of mobile GPU/AI chips, and reduce CPU usage.

3. Easy integration and high compatibility

Provide standardized Android API interfaces (Java/Kotlin dual language support), along with complete development documentation, demo projects, and debugging tools. Developers can complete integration debugging in 1-2 days;

Compatible with Android 8.0 to the latest Android 14 version, supports 64 bit/32-bit applications, adapts to irregular screens such as notch screens and hole digging screens, and is compatible with mainstream Android development frameworks such as Jetpack.

4. Safety and Compliance

Local facial data processing is completed without uploading to the cloud, avoiding user privacy leakage and complying with regulatory requirements such as the Personal Information Protection Law and the Data Security Law;

Support live detection (silent/action), which can effectively resist attack methods such as photos, videos, 3D masks, etc. The accuracy of live detection is ≥ 99.5%;

Provide data encryption interface to desensitize facial feature data and meet financial level security standards.

5. Functional modularization

Adopting modular design, core functions can be integrated as needed (such as only integrating 3D face detection or simultaneously integrating reconstruction and liveness detection), reducing the size of application packages;

Support custom parameter configuration (such as detection threshold, reconstruction accuracy) to adapt to personalized needs in different scenarios.

1. Limitations of traditional 2D facial technology

To solve the problem of 2D face detection being easily affected by angles, lighting, and occlusion, and unable to restore the true 3D structure of the face, and to achieve more accurate facial feature extraction;

Resolve the issue of 2D liveness detection being easily deceived by photos and videos, and enhance the security of identity verification;

Resolve the issue of 2D facial applications being unable to adapt to 3D scenes such as AR makeup, 3D facial printing, and virtual image customization, and expand the application boundaries.

2. The technical threshold issue of 3D face development on mobile devices

To solve the problem of lack of 3D computer vision and deep learning algorithm research and development capabilities for enterprises/developers, high development costs and long cycles from scratch, and to quickly implement applications through SDK out of the box capabilities;

To solve the problem of algorithm adaptation and poor compatibility caused by mobile hardware fragmentation (large differences in camera parameters and computing power among different models), the SDK has completed full model adaptation to reduce adaptation costs;

To solve the problems of limited mobile resources (memory, computing power, storage), stuttering and high power consumption of 3D algorithms, the SDK has been optimized for lightweight design, balancing performance and user experience.

3. Actual pain points in industry scenarios

In the field of finance/security: solve the problems of facial fraud and identity impersonation in remote identity verification, and achieve high security level identity verification through 3D liveness detection;

In the field of beauty/AR, we aim to solve the problem of poor fit between 2D makeup attempts and real faces. Based on 3D face reconstruction, we aim to achieve precise virtual makeup and wearing (glasses/jewelry);

In the field of medical beauty, we solve the problem of relying on professional equipment for facial contour measurement and medical beauty effect simulation. We use mobile 3D facial scanning to quickly generate 3D facial data and assist in the design of medical beauty solutions;

In the social/entertainment field, we solve the problem of low matching between virtual images and real human faces by generating highly restored personalized virtual images based on 3D facial features;

In the field of attendance/access control, we aim to solve the problem of low facial recognition accuracy in complex environments such as backlighting and wearing masks, and improve the stability of attendance/access control systems.

4. Data security and compliance issues

Resolve privacy leakage risks caused by cloud based transmission and storage of facial data, and meet data compliance requirements through local processing;

The SDK provides data encryption and anonymization capabilities that meet regulatory requirements to address the issue of facial data not being desensitized and being easily abused in industry applications.