Content Detail

Camera2 YUV Input Review: Strides and Frame Lifetime in a Vision SDK

Camera2 YUV_420_888 input requires plane-aware rowStride and pixelStride handling, orientation rules and frame lifetime management. This review supplies an integration checklist based on an existing SDK case.

Industry Updates 2026-10-17 Winge Technology
Article Content

Do not assume a Camera2 buffer is contiguous NV21 based only on width and height. Read according to the image format and each plane's rowStride and pixelStride. Define rotation, mirroring and when asynchronous processing finishes. This is a project and documentation review, not an Android release announcement.

What the documentation establishes

Android's ImageFormat reference and Image.Plane API describe YUV_420_888 and plane strides. Read values returned by the target device instead of assuming another phone or board has the same layout. No additional device compatibility testing is reported here.

SDK input checklist

Item Inputs to record Failure checks
Format and planes Format, dimensions, plane strides and crop region Use edge-pattern samples to check row errors, color errors and out-of-bounds access
Rotation and mirroring Sensor orientation, display orientation and front-camera mirroring Check portrait/landscape and left/right; test coordinate mapping separately
Lifetime Responsibility for retaining, copying, submitting and releasing Image data Asynchronous work must not read released or invalid buffers
Queue and load Capture rate, submission rate, backlog and dropping policy Record latency and buffer release during sustained input
Output coordinates Transform between model input and preview region Check boxes against samples at known positions

Organizing integration validation

Start with one specified device and input mode. Save input metadata and publishable samples, then test conversion, orientation and coordinates. Add other resolutions and sustained capture afterward. Exclude conversion and lifetime errors before attributing changed results to the algorithm.

Published project boundaries

The published vision SDK project record in Chinese covers offline face detection, orientation assessment and Camera2 integration on RK3588 Android. Detection is not identity recognition. The CPU path does not establish NPU acceleration or compatibility with all Android devices. Provide target hardware, system, samples and SDK requirements when assessing the original Chinese Android development service, then agree a device matrix.

Submit a Project Requirement

Leave your contact details and a brief requirement so we can review the technical direction and scope.

Online
Phone
13910119357
WeChat
WhatsApp
Winge Technology WhatsApp QR code Scan or click to contact us
Top