After successful AAR integration, verify that actual devices load dependencies and models, process camera input and return corresponding results correctly. A successful build supports only that build combination. Installation, preview and detection need separate evidence. They do not establish acceptance of the entire device.
Which device conditions should be checked first?
Record device, system, architecture, SDK and model versions. Check runtime dependencies, permissions and resource installation. See official Android library documentation for the AAR format. Use delivery inventories and actual runtime records for specific dependencies and support scope.
How should camera input and display be verified?
Save actual input frames. Check format, dimensions, strides and rotation, then compare result coordinates and preview overlays. Test mirroring, cropping, camera switching, stop and restart. Ensure old-frame results or invalid buffers are not treated as current valid outputs. If only one example image works, state that the actual camera pipeline remains unverified.
Can example performance and accuracy results be reused?
Not directly. Define measurements for target devices, input modes, models, threads and workloads. Record submission timing, processing time and display delay separately. Accuracy needs agreed samples, labels, statistics and failure states. Long-duration operation, electrical power consumption and temperature rise need corresponding test records. An example submission interval or processor specification is not a measured result for those metrics.
What else is needed for formal delivery?
Confirm known issues, recovery steps, resource usage scope, signed versions, updates and acceptance conditions separately. A customer-test package or Debug example does not automatically become a production-signed package. This article describes checks. It reports no new passed devices, recognition rate or production status.
Related services: Android, mobile and embedded vision SDKs, project consultation.