HRNetPoseQuantized: Optimized for Mobile Deployment

Perform accurate human pose estimation

HRNet performs pose estimation in high-resolution representations.

This model is an implementation of HRNetPoseQuantized found here.

This repository provides scripts to run HRNetPoseQuantized on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

Model Type: Pose estimation
Model Stats:
- Model checkpoint: hrnet_posenet_FP32_state_dict
- Input resolution: 256x192
- Number of parameters: 28.5M
- Model size: 109 MB

Model	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Precision	Primary Compute Unit	Target Model
HRNetPoseQuantized	Samsung Galaxy S23	Snapdragon® 8 Gen 2	TFLITE	0.95 ms	0 - 22 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	Samsung Galaxy S23	Snapdragon® 8 Gen 2	QNN	1.24 ms	0 - 11 MB	INT8	NPU	HRNetPoseQuantized.so
HRNetPoseQuantized	Samsung Galaxy S24	Snapdragon® 8 Gen 3	TFLITE	0.706 ms	0 - 39 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	Samsung Galaxy S24	Snapdragon® 8 Gen 3	QNN	0.915 ms	0 - 34 MB	INT8	NPU	HRNetPoseQuantized.so
HRNetPoseQuantized	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	TFLITE	0.645 ms	0 - 36 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite	QNN	0.888 ms	0 - 32 MB	INT8	NPU	Use Export Script
HRNetPoseQuantized	RB3 Gen 2 (Proxy)	QCS6490 Proxy	TFLITE	3.735 ms	0 - 44 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	RB3 Gen 2 (Proxy)	QCS6490 Proxy	QNN	5.349 ms	0 - 7 MB	INT8	NPU	Use Export Script
HRNetPoseQuantized	RB5 (Proxy)	QCS8250 Proxy	TFLITE	17.373 ms	0 - 3 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	QCS8550 (Proxy)	QCS8550 Proxy	TFLITE	0.954 ms	0 - 11 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	QCS8550 (Proxy)	QCS8550 Proxy	QNN	1.218 ms	0 - 2 MB	INT8	NPU	Use Export Script
HRNetPoseQuantized	SA7255P ADP	SA7255P	TFLITE	13.862 ms	0 - 33 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	SA7255P ADP	SA7255P	QNN	14.372 ms	0 - 10 MB	INT8	NPU	Use Export Script
HRNetPoseQuantized	SA8255 (Proxy)	SA8255P Proxy	TFLITE	0.962 ms	0 - 19 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	SA8255 (Proxy)	SA8255P Proxy	QNN	1.223 ms	0 - 1 MB	INT8	NPU	Use Export Script
HRNetPoseQuantized	SA8295P ADP	SA8295P	TFLITE	1.675 ms	0 - 36 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	SA8295P ADP	SA8295P	QNN	1.997 ms	0 - 6 MB	INT8	NPU	Use Export Script
HRNetPoseQuantized	SA8650 (Proxy)	SA8650P Proxy	TFLITE	0.951 ms	0 - 16 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	SA8650 (Proxy)	SA8650P Proxy	QNN	1.211 ms	0 - 1 MB	INT8	NPU	Use Export Script
HRNetPoseQuantized	SA8775P ADP	SA8775P	TFLITE	1.455 ms	0 - 36 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	SA8775P ADP	SA8775P	QNN	1.913 ms	0 - 6 MB	INT8	NPU	Use Export Script
HRNetPoseQuantized	QCS8450 (Proxy)	QCS8450 Proxy	TFLITE	1.216 ms	0 - 43 MB	INT8	NPU	HRNetPoseQuantized.tflite
HRNetPoseQuantized	QCS8450 (Proxy)	QCS8450 Proxy	QNN	1.481 ms	0 - 39 MB	INT8	NPU	Use Export Script
HRNetPoseQuantized	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN	1.347 ms	0 - 0 MB	INT8	NPU	Use Export Script

Installation

This model can be installed as a Python package via pip.

pip install "qai-hub-models[hrnet_pose_quantized]"

Configure Qualcomm® AI Hub to run this model on a cloud-hosted device

Sign-in to Qualcomm® AI Hub with your Qualcomm® ID. Once signed in navigate to Account -> Settings -> API Token.

With this API token, you can configure your client to run models on the cloud hosted devices.

qai-hub configure --api_token API_TOKEN

Navigate to docs for more information.

Demo off target

The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input.

python -m qai_hub_models.models.hrnet_pose_quantized.demo

The above demo runs a reference implementation of pre-processing, model inference, and post processing.

NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).

%run -m qai_hub_models.models.hrnet_pose_quantized.demo

Run model on a cloud-hosted device

In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following:

Performance check on-device on a cloud-hosted device
Downloads compiled assets that can be deployed on-device for Android.
Accuracy check between PyTorch and on-device outputs.

python -m qai_hub_models.models.hrnet_pose_quantized.export

Profiling Results
------------------------------------------------------------
HRNetPoseQuantized
Device                          : Samsung Galaxy S23 (13)
Runtime                         : TFLITE                 
Estimated inference time (ms)   : 0.9                    
Estimated peak memory usage (MB): [0, 22]                
Total # Ops                     : 518                    
Compute Unit(s)                 : NPU (518 ops)

Run demo on a cloud-hosted device

You can also run the demo on-device.

python -m qai_hub_models.models.hrnet_pose_quantized.demo --on-device

NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).

%run -m qai_hub_models.models.hrnet_pose_quantized.demo -- --on-device

Deploying compiled model to Android

The models can be deployed using multiple runtimes:

TensorFlow Lite (.tflite export): This tutorial provides a guide to deploy the .tflite model in an Android application.
QNN (.so export ): This sample app provides instructions on how to use the .so shared library in an Android application.

View on Qualcomm® AI Hub

Get more details on HRNetPoseQuantized's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of HRNetPoseQuantized can be found here.
The license for the compiled assets for on-device deployment can be found here

References

Community

Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
For questions or feedback please reach out to us.