Configure sample applications - Qualcomm Dragonwing Documentation

This section provides a flexible workflow for running and customizing advanced multimedia and AI sample applications using the Qualcomm® Intelligent Multimedia Product (QIMP) SDK on Ubuntu. Developers can define input/output sources, runtime targets, and model precision using JSON configuration files—enabling seamless evaluation across CPU, GPU, and DSP. With support for frameworks like TFLite, QNN, and SNPE, and integration with AI Hub, this setup is ideal for building and optimizing edge AI pipelines tailored to specific use cases. Running advanced multimedia and AI sample applications using the QIMP SDK on Ubuntu enables developers to:

Prototype and validate AI workloads across heterogeneous compute targets (CPU, GPU, DSP), helping teams choose the most efficient runtime for their use case.
Customize application behavior using JSON-based configuration, allowing precise control over input/output sources, model types, and runtime parameters.
Accelerate development and deployment by leveraging pre-integrated models from AI Hub and supported frameworks like TFLite, QNN, and SNPE.
Benchmark performance and optimize resource usage, which is critical for embedded systems and edge devices where compute and power budgets are constrained.
Ensure compatibility and reproducibility across Qualcomm platforms by using standardized scripts and directory structures for models, labels, and media assets.

To begin configuring sample applications on your Dragonwing device, follow these step-by-step instructions to install required dependencies and configure the system. Prerequisites

Setup the Device.
Ensure that the IQ‑9075 device is running Ubuntu OS.
Setup Wi-Fi & access IQ-9075 EVK device via SSH.
Basic familiarity with JSON configuration files and runtime environment variables.
Access to AI Hub for model selection and export.
- Create AI Hub account** if you don’t already have one**
If the PPA packages are not already installed, run the Install needed packages step.

Install needed packages

To install the needed pre-built packages, refer to the 🔗Install required software packages section and run the provided “Install PPA Packages” script.

Download and run the artifact script

Before running any sample applications, you must download the required models, labels, and media assets. The download_artifacts.sh script automatically fetches all required components, including:

Models – Pretrained machine‑learning models used by the sample applications for inference.
Labels – Metadata files that map model outputs to human‑readable class names.
Media files – Sample input data (such as images or videos) used to run and validate the applications.

You can run the following commands via SSH, or logged on locally to your device.

cd /home/ubuntu 
curl -L -O https://raw.githubusercontent.com/quic/sample-apps-for-qualcomm-linux/refs/heads/main/qualcomm-linux/scripts/download_artifacts.sh
sudo chmod +x download_artifacts.sh 
sudo ./download_artifacts.sh

Verify Model/Label/Media Files

Before launching any sample applications, make sure the required files are in place.

Check the following directories:

Model files → /etc/models/
Label files → /etc/labels/
Media files → /etc/media/

Note: These files are essential for AI sample applications to function correctly. If they’re missing, re-run the artifact download script.

Updating JSON Config File

To run sample applications with a specific functionality, you’ll need a properly configured JSON file.

What to Do

Update the required JSON config file based on your model and config requirements.

Configuration Parameters

parameter	Option
Input Source	Camera, File (Filesrc), RTSP Stream
Output Source	Waylandsink, Filesink, RTSP Stream
Runtime Options	CPU, GPU, DSP
Precision	INT8 or INT16, W8A8 or W8A16, FP32
Model Type	Select from available models in AI Hub
Labels	Select the correct labels file

Example 1:
The gst-ai-multistream-inference application shows AI inference (object detection and classification) on up to 32 input streams coming from camera, file, or RTSP streams. To run the multistream inference sample application for object detection using 32 H.264 file inputs and display the output on a Wayland display, configure the input file paths to reference /etc/media/video.mp4 for all 32 streams. Open the corresponding config-multistream-inference.json file and apply the required changes.

sudo vi /etc/configs/config-multistream-inference.json

{
  "input-file-path":
    [
      "/etc/media/video.mp4",
      "/etc/media/video.mp4",
      "/etc/media/video.mp4",
      "/etc/media/video.mp4",
      .......
       "/etc/media/video.mp4",
    ]
}

Example 2:
The gst-ai-object-detection application allows you to detect objects within images and videos. To consider the input source from a live camera feed, configure the input source as

#Primary
"camera":0
#Secondary
"camera":1

Run the sample application

To run the sample application, do the following:

Use the SSH or SBC terminal to launch the sample application.
Set the required environment variables to connect a Wayland application to the active user session.

export XDG_RUNTIME_DIR=/run/user/$(id -u ubuntu)
export WAYLAND_DISPLAY=wayland-1

For Desktop build:
Run the following additional command to creates a missing symbolic link (for TensorFlow Lite C library).

sudo ln -s /lib/aarch64-linux-gnu/libtensorflowlite_c.so.2 /lib/aarch64-linux-gnu/libtensorflowlite_c.so

Run sample application.\

gst-ai-multistream-inference

gst-ai-multistream-inference pipeline:

Expected Output:

To display the available help options, run the following command in the SSH shell:

gst-ai-multistream-inference -h

For Debug Messages:
Set GST_DEBUG environment variable used by GStreamer to control the verbosity and category of debug messages.

export GST_DEBUG=3

To stop the application, press CTRL + C

Sample Application Configuration Matrix

Sample App Name	Details	AI Hub Model Type	Runtime	Script to Use
gst-ai-classification	Image classification	MobileNet-v2, ResNet101, GoogLeNet, MobileNet-v3-Large, ResNet18, ResNeXt50, ResNeXt101, SqueezeNet, WideResNet50, Shufflenet	CPU, GPU, DSP	Update JSON
gst-ai-object-detection	Object detection	Yolox, Yolov7, Yolov8-Detection (manual export)	CPU, GPU, DSP	Export model from AI Hub; Update script for Yolox/Yolov7 – Update JSON
gst-ai-pose-detection	Pose detection	hrnet_pose	CPU, GPU, DSP	TFLite works by default; update script for precision/runtime – Update JSON
gst-ai-segmentation	Image segmentation	FFNet-40S, FFNet-54S, FFNet-78S	CPU, GPU, DSP	Update JSON
gst-ai-superresolution	Video super-resolution	quicksrnetsmall, QuickSRNetMedium, QuickSRNetLarge, XLSR	CPU, GPU, DSP	Update JSON
gst-ai-multistream-batch-inference	Multistream batch inference	YoloV8-Detection (batch 4), DeeplabV3 (batch 4)	CPU, GPU, DSP	Export model from AI Hub; Update script – Update JSON
gst-ai-face-detection	Face detection	face_det_lite	CPU, GPU, DSP	Update JSON
gst-ai-face-recognition	Face recognition	face_det_lite, face_attrib_net, facemap_3dmm	CPU, GPU, DSP	Face registration required; otherwise output is ‘unknown face recognized’
gst-ai-metadata-parser-example	Metadata parsing	Yolov8-Detection	CPU, GPU, DSP	Export model from AI Hub
gst-ai-usb-camera-app	AI USB camera	Yolov8-Detection	CPU, GPU, DSP	Export model from AI Hub
gst-ai-parallel-inference	Parallel inferencing	Yolov8-Detection, Deeplab, Hrnet, Inceptionv3	CPU, GPU, DSP	Export model from AI Hub; Update JSON for other models
gst-ai-daisychain-detection-classification	Daisy chain detection and classification	Inceptionv3 + YoloV8	CPU, GPU, DSP	Export model from AI Hub; Update JSON for other models
gst-ai-audio-classification	Audio classification	Inceptionv3 + YoloV8	CPU, GPU, DSP	Export model from AI Hub; Update JSON for other models
gst-ai-smartcodec-example	AI smart codecn	Inceptionv3 + YoloV8	CPU, GPU, DSP	Export model from AI Hub; Update JSON for other models

Reference Docs

To further explore sample applications, see the Qualcomm Intelligent Multimedia SDK (IM SDK) Reference Guide. Qualcomm Intelligent Multimedia SDK (IM SDK) Reference

​Install needed packages

​Download and run the artifact script

​Verify Model/Label/Media Files

​Check the following directories:

​Updating JSON Config File

​What to Do

​Configuration Parameters

​Run the sample application

​Sample Application Configuration Matrix

​Reference Docs

Install needed packages

Download and run the artifact script

Verify Model/Label/Media Files

Check the following directories:

Updating JSON Config File

What to Do

Configuration Parameters

Run the sample application

Sample Application Configuration Matrix

Reference Docs