The published JIVISION project for RK3588 smart glasses documents Android offline face detection and a delivery for customer testing. This review examines module organization in the published case and publication record. It adds no device tests. This review did not unpack any AAR. It did not recalculate the original delivery-package checksums.
Publication record date: 2026-07-31. Original SDK module version: 0.4.5.
Check both modules with external resources
| Item | Role or content in the original published case | Review limit |
|---|---|---|
Core AAR |
Common runtime, states and basic utilities | Names do not establish implementation in other versions |
Face AAR |
Face-detection interfaces, configuration and result structures | No person-identity confirmation result |
| OpenCV and model | Specified dependency; processing route: YuNet CPU |
No inferred compatibility for other acceleration backends or devices |
| Examples and integration records | Demo, source code, model, guides and status notes | Examples and documents do not replace customer-environment acceptance |
Version relationships come from the original record
The original published case requires both AARs to use the same SDK version with the specified dependencies and model. Recorded dependency: OpenCV 4.10.0. Verified architecture: arm64-v8a. Recorded processing route: YuNet CPU. These describe the original delivery combination, not a compatibility inventory for every Android system, camera or acceleration backend.
Preserve customer-test status
The original case limits delivery to customer functional testing and Android application integration assessment. It is not a production-signed version. Original build, package-integrity or test results do not replace accuracy checks on agreed customer samples, long-duration stability, actual-device electrical power consumption and temperature-rise measurements, or formal acceptance. This review does not report those checks as newly passed results.
Separate handover checks from extensions
Further work can associate AARs, dependencies, models, examples and instructions by version, checking installation, resource loading, result fields and known limits. This is a review recommendation. It does not report a newly obtained package inventory or device-verification record. The original scope remains face detection. It adds no identity registration, feature extraction, identity-database comparison, name or document recognition, and makes no production-readiness claim.
Fact source: RK3588 smart-glasses offline-detection case and its publication record. Related services: Android, mobile and embedded vision SDKs, project consultation.