API coverage

Find examples by feature and API. Start with the platform guide to install and initialize the SDK, then choose your API in each feature guide's code tabs.

The examples use InspireFace 1.2.4, InspireCV 1.0.2, Android Java SDK 1.2.4.post1, and HarmonyOS ArkTS SDK 1.2.4. Python uses the 1.2.4.post3 PyPI package, including the 1.2.4 native SDK. Objective-C and Swift use the 1.2.4 Apple package; the HarmonyOS release includes its native SDK and HAR project. Java (JVM) uses the portable JNI build from the 1.2.4 source and is listed separately from Android. Windows x64 uses the C, C++ and Python columns below; it has no separate API layer. The table lists the operations available through each interface.

Feature examples by API

“Example” links to a guide with code for that API. The Android 1.2.4.post1 AAR includes both the InspireFace facade and the complete jni.Native API. Its column covers both, with each example identifying the layer it uses. A dash means the named interface does not directly expose that operation.

Feature	C API	C++	Java	Android	Python	HarmonyOS (ArkTS)	Objective-C	Swift
Launch / Session / release	Example	Example	Example	Example	Example	Example	Example	Example
Detection / tracking	Example	Example	Example	Example	Example	Example	Example	Example
Dense landmarks	Example	Example	Example	Example	Example	Example	Example	Example
Five-point landmarks	Example	Example	Example	Example	Example	Example	Example	Example
Quality / mask / attributes	Example	Example	Example	Example	Example	Example	Example	Example
Pose	Example	Example	Example	Example	Example	Example	Example	Example
Expression	Example	Example	Example	Example	Example	Example	Example	Example
RGB liveness / actions	Example	Example	Example	Example	Example	Example	Example	Example
Embedding / comparison	Example	Example	Example	Example	Example	Example	Example	Example
FeatureHub	Example	Example	Example	Example	Example	Example	Example	Example
Alignment crop	Example	Example	Example	Example	—	Example	Example	Example
Aligned-image extraction	Example	Example	Example	Example	—	Example	Example	Example
Similarity display conversion	Example	Example	Example	Example	Example	Example	Example	Example
Detection snapshots	Example	Value copy	Example	Example	Example	Example	Example	Example
Face capture	Example	Example	Example	Example	Example	Example	Example	Example
CPU power mode	—	Example	Example	Example	—	—	—	—

The Plus passive and flash demos combine mobile capture with service-side verification. Their guide covers Android capture, progress feedback and result handling.

The code tabs share the same API selector, so choosing Python, Java, Android, HarmonyOS (ArkTS), Objective-C or Swift also selects it in other groups that contain the same option. The complete examples include full programs to copy from the page; snippets in feature guides state which session, image or model setup they expect.

Inputs, settings and deployment

Area	Entry points	Guide
SDK downloads and builds	Prebuilt packages, platform builds and Python native-library replacement	Get and build the SDK, Python packaging
Model loading and inspection	Launch, reload, resource-pack validation and metadata	Models and builds
Images and buffers	File/bitmap input, raw RGB/BGR/YUV, stream updates, strides and rotation	Image inputs
Session tuning	Detector level, minimum face size, confidence, preview size, intervals and smoothing	Tracking
Memory and threads	Borrowed buffers, owned results, session reuse and worker teardown	Architecture
C ABI	Handles, status codes and `HFSessionConfigV2`	C API
Java deployment	Java 8 JAR, JNI paths, process architecture, direct buffers and explicit release	Java, Java packaging
Android deployment	1.2.4.post1 AAR, single native library per ABI, R8, CameraX and both Java API layers	Android
Windows deployment	x64 CPU, C/C++ CMake integration, Python wheels and DLL runtime setup	Windows, Windows builds
Apple APIs	Objective-C errors, Swift types, scoped views and complete detection examples	Objective-C and Swift
iOS deployment	Device / simulator XCFrameworks, Xcode linkage, camera buffers and CoreML	iOS
macOS deployment	Intel / Apple Silicon frameworks, embedding, signing and native libraries	macOS
HarmonyOS	HAR, ArkTS, Node-API and worker ownership	HarmonyOS
ARM CPU	Image preprocessing, memory reuse and camera latency	ARM deployment
NVIDIA	TensorRT SDK, CUDA runtime and device selection	TensorRT
Rockchip	SoC model packs, toolchains, RK runtime and RGA	Rockchip, Python on Rockchip
Diagnostics	Version, error text, runtime diagnostics and resource counters	API recipes, Troubleshooting
Performance	Warm-up, timing boundaries, median/p95 and queue delay	Performance

For parameter types, overloads and additional settings, refer to the headers and wrapper shipped with your SDK version. A Java installation includes Native.java, NativeTypes.java and NativeConstants.java under sources/, plus api-manifest.json listing generated functions. Every C API has a Java mapping; hardware-specific calls still depend on the native build options and runtime.

Image processing with InspireCV

InspireCV is a standalone C++ library. Its guide covers Image I/O and operations, float images, Task transforms and tensors, error handling, repeated-frame memory reuse and optional CUDA processing. Its PixelFormat, rotation and pipeline types are separate from InspireFace's C API and FrameProcess.

For the full public interface, see InspireFace C declarations, native C++ headers, Objective-C header, Swift overlay, Python wrapper, HarmonyOS ArkTS exports and InspireCV headers. Use the headers bundled with a release as the reference for its binary.

Integration notes

Java long handles require explicit release. A result’s ByteBuffer can borrow native storage, so retaining the Java object does not keep that storage valid. See Java lifetimes.
Android 1.2.4.post1 includes capture and snapshots. Facade results own copied Java arrays and tokens; Native buffers retain C lifetime rules. Release streams through the API layer that created them. See Android lifetimes.
Copying a C++ face-result vector retains its geometry and tokens. Keep the corresponding frame pixels as well if you will extract features later.
In ArkTS, release tracking snapshots with session.releaseFaceResult() and call close() on streams, bitmaps, capture sessions and sessions when finished. Keep each wrapper object on the worker that created it.
The standard HarmonyOS HAR runs MNN on CPU and accepts raw image buffers. Use HarmonyOS APIs to decode files and display images.
Objective-C and Swift wrap the native handles. ARC owns the wrapper object; borrowed face, feature and pixel pointers still have the native lifetime limits.
CoreML, TensorRT, RKNN, RGA and CUDA preprocessing require their matching SDK build and runtime. Follow the target platform's deployment guide.