Qt Edge AI prerequisites

Qt Edge AI runs AI inference on the device. Before you install it, make sure that you have supported hardware, an AI model for your use case, and the third-party software that the inference backend needs.

Supported platforms

Qt Edge AI Technical Preview 1.0 supports the following platforms. Each platform has its own operating system and AI runtime.

Other platforms are not supported. Qt Edge AI builds on a development PC, but it runs inference only on the supported platforms.

The NVIDIA devices are tested with JetPack 7.x. The Qualcomm devices have been tested with the Qt Onboard 6.12 release.

Hardware

You need one of the following devices:

• A Qualcomm Dragonwing IQ-9075 Evaluation Kit (EVK)
• An NVIDIA Jetson AGX Orin Developer Kit
• An NVIDIA Jetson AGX Thor Developer Kit

You also need:

• A USB camera, for applications that use video input. CSI and MIPI cameras are not supported. The Simple Face Detection example needs a camera. Tested cameras include Logitech C922 Pro HD Stream Webcam.
• A display, for applications that show video. The Simple Face Detection example shows the camera video in a window.
• A network connection to the device, to install packages and to deploy applications.

AI models

Qt Edge AI does not include AI models. You must provide a model that fits your use case and the AI runtime of your platform. You name the model in the configuration file of the VisionInference step, which the step's configFile property sets. See Qt Edge AI Configuration Files.

AI models for DeepStream

The Simple Face Detection example uses the primary detector model of the DeepStream sample applications. The DeepStream SDK installs it in /opt/nvidia/deepstream/deepstream/samples/models/Primary_Detector/, so there is nothing to download. The model detects cars, bicycles, people and road signs, not faces. See DeepStream sample applications.

AI models for Edge Impulse

Register to Edge Impulse and create your own AI model for face detection, e.g. in Face detection - FOMO. Follow the documentation of the Edge Impulse. Readymade AI models are not available from Edge Impulse so you need to train your own model and deploy it for your platform. On Qualcomm Dragonwing IQ9, deploy it for the Qualcomm IM SDK GStreamer plugin (AARCH64, QNN).

AI models for Qualcomm IM SDK

For IM SDK the supported AI runtime is TFLite. You can find pre-trained and converted models from Qualcomm AI Hub, e.g. for face detection. By selecting the TFLite and Dragonwing IQ-9075 EVK you can download the right AI model. The example expects the file name Lightweight-Face-Detection_w8a8.tflite. Change the model key in the configuration file if your file has another name. See Installing Qt Edge AI with a QBSP image.

Third-party software

The following third-party software must be present on the device. Other versions might work, but Qt does not test them.

Supported Qt versions

Qt Edge AI Technical Preview 1.0 is based on Qt 6.12. It requires Qt 6.12.0 or later. Earlier Qt versions, such as Qt 6.11, are not supported.

Qt Edge AI uses the Qt Core, Qt Qml, Qt Quick, and Qt Multimedia modules.

The imsdk step provider needs the camera frames in DMA buffers. Qt Multimedia passes them on only when it is built with -DFEATURE_gstreamer_qt_api=ON and runs with QT_MEDIA_BACKEND set to gstreamer. Otherwise the IM SDK step logs a warning and never becomes ready.

Licensing

• Qt Edge AI is available under a commercial license only. You need a Qt for Device Creation license, or an evaluation license.
• You need a Qt Account to access the installers and package repositories.

Next steps

After you have the prerequisites, install Qt Edge AI. For the available installation methods, see Installing Qt Edge AI.