Note: README is still a work in progress
This app is provided as a pose detection sample using the open source HRNET-POSE model from Qualcomm AI Hub. The application uses ONNX runtime (ORT) to enable the model to run cross-platform.
On the Snapdragon X Elite, the model is optimized to leverage the Neural Processing Unit (NPU) at inference runtime. Elsewhere, it will run using the CPU.
This sample python app has only been validated using Windows 11 Enterprise Snapdragon(R) X Elite
Follow these steps to setup the app for your platform.
-
git clone repo
-
Create virtual environment
>> python3.11 -m virtual_env env_sample_app_hrnet
-
Activate virtual environment
>> env_sample_app_hrnet/Scripts/activate.ps1 (Windows: Validated) >> src env_sample_app_hrnet/bin/activate (Mac: Validated) >> src env_sample_app_hrnet/bin/activate (Linux: Not Validated) # Will not work via WSL due to camera binding issue within WSL
-
Install dependencies
>> pip install -r requirements.txt
-
Download model from AI Hub https://aihub.qualcomm.com/compute/models/hrnet_pose?domain=Computer+Vision&useCase=Pose+Estimation
-
Transfer model to qnn_sample_apps/models/
>> mv Downloads/hrnet_pose.onnx qnn_sample_apps/models/
Coming Soon
>> python ./src/hrnet_pose/main.py (from root directory)
>> python ./src/hrnet_pose/main.py --system windows --model hrnet_pose --processor cpu --camera 1 --available_cameras False
We welcome contributions to this repository! Please refer to our contributing guide for how to contribute.
qnn_sample_apps
├─ .gitignore
├─ dll.json
├─ executioner.json
├─ models
│ └─ README.md
├─ models.json
├─ notebooks
│ └─ sample_app_hrnet.ipynb
├─ pyproject.toml
├─ README.md
├─ requirements.in
├─ requirements.txt
├─ scripts
│ └─ directory_information.txt
├─ setup.py
├─ src
│ ├─ hrnet_pose
│ │ ├─ main.py
│ │ ├─ model_inference.py
│ │ ├─ model_loader.py
│ │ └─ README.md
│ └─ __init__.py
└─ tests
└─ test_module.py