MediaPipe-Hand-Detection Example

This document describes how to use the QAI AppBuilder Python API to run inference with the MediaPipe-Hand-Detection hand pose recognition model on Qualcomm® Hexagon™ Processor (NPU).

Supported devices

Device	SoC
Fogwise® AIRbox Q900	QCS9075

Install QAI AppBuilder

tip

1. Install QAI AppBuilder by following the QAI AppBuilder installation guide.

2. Configure ADSP environment variables as described in Create ADSP environment variables.

Run the sample

Install dependencies

Install sample dependencies in the activated virtual environment:

Device

pip3 install requests tqdm qai-hub py3-wget Pillow torch torchvision opencv-python-headless pygame

The script also looks under ~/.qaihm for MediaPipePyTorch (provides blazepalm and related modules). If missing:

Device

mkdir -p ~/.qaihm
git clone --depth 1 https://github.com/zmurez/MediaPipePyTorch.git ~/.qaihm/MediaPipePyTorch

Run the script

• Enter the upstream samples directory

  Device

  cd qai-appbuilder/samples

• Prepare input data (camera UI needs a GUI; use a static image in headless environments)

input image

• Run inference

  Device

  python3 ComputerVision/Pose_Estimation/mediapipe_hand/mediapipe_hand.py \
    --chipset 9075 \
    --imagefile /path/to/hand_image.png \
    --displayPredict False \
    --DemoMode False

• Example result: output.jpg is written in the sample directory

output image

tip

The first run downloads handdetector and landmarkdetector via Qualcomm® AI Hub. Prefer --imagefile when no display is available, and use opencv-python-headless to avoid Qt plugin initialization failures.

python3 run_inference.py --list
python3 run_inference.py --model mediapipe_hand --args "--chipset 9075 --imagefile /path/to/hand_image.png --displayPredict False --DemoMode False"