Qt Edge AI C++ Classes

Detailed Description

The Qt Edge AI C++ classes describe an inference pipeline as a flow of steps that exchange data through typed ports, and provide the interfaces that backend plugins implement to run the pipeline.

To use these classes in your application, add the following include statement:

#include <QtEdgeAi/QtEdgeAi>

Example

The following builds a minimal flow: a camera feeds a vision inference step, whose output is parsed into JSON.

QCamera camera;
QMediaCaptureSession session;
session.setCamera(&camera);
camera.start();

auto *videoSource = new QInferenceVideoSource(&session);
auto *detection = new QVisionInference(QUrl("configs/edgeimpulse/inference.json"), "edgeimpulse");
auto *parser = new QInferenceResultParser;

QInferenceFlow *flow = QInferenceFlow::fromSteps({ videoSource, detection, parser });
QObject::connect(parser, &QInferenceStep::dataAvailable,
                  [](const QVariant &data, const QString &) { qDebug() << data; });
flow->start();

The vision step reads its configuration from configs/edgeimpulse/inference.json, relative to the directory of the application's executable. The file names the model to run, by the file name that Edge Impulse gives the library. A renamed library fails to load:

{
    "pluginPath": "libgstedgeimpulse-898869-face-detection-fomo.so"
}

The model is not shipped with the module. Two sources are common: a model already optimized for the target hardware, such as a .tflite model from the Qualcomm Intelligent Multimedia SDK's model zoo, or a model trained on your own data through Edge Impulse Studio, which exports it as a compiled GStreamer plugin like the one above. Either way, the step's configFile gives the path to the resulting file, and the step's provider selects the step provider that loads it: edgeimpulse for an Edge Impulse plugin, imsdk for a Qualcomm IM SDK model. See Qt Edge AI Configuration Files for what each step provider reads.