LiteHRNet: Optimized for Qualcomm Devices

LiteHRNet is a machine learning model that detects human pose and returns a location and confidence for each of 17 joints.

This is based on the implementation of LiteHRNet found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.

Getting Started

There are two ways to deploy this model on your device:

Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

Runtime	Precision	Chipset	SDK Versions	Download
ONNX	float	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.19.1	Download

For more device-specific assets and performance metrics, visit LiteHRNet on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

Use the Qualcomm® AI Hub Models Python library to compile and export the model with your own:

• Custom weights (e.g., fine-tuned checkpoints)
• Custom input shapes
• Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for LiteHRNet on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.pose_estimation

Model Stats:

• Input resolution: 256x192
• Number of parameters: 1.11M
• Model size (float): 4.49 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
LiteHRNet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	2.744 ms	1 - 100 MB	NPU
LiteHRNet	ONNX	float	Snapdragon® X2 Elite	2.854 ms	5 - 5 MB	NPU
LiteHRNet	ONNX	float	Snapdragon® X Elite	5.72 ms	5 - 5 MB	NPU
LiteHRNet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	3.143 ms	0 - 129 MB	NPU
LiteHRNet	ONNX	float	Qualcomm® QCS8550 (Proxy)	5.199 ms	0 - 14 MB	NPU
LiteHRNet	ONNX	float	Qualcomm® QCS9075	5.909 ms	1 - 4 MB	NPU
LiteHRNet	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	2.845 ms	0 - 98 MB	NPU
LiteHRNet	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.833 ms	1 - 85 MB	NPU
LiteHRNet	QNN_DLC	float	Snapdragon® X2 Elite	1.23 ms	1 - 1 MB	NPU
LiteHRNet	QNN_DLC	float	Snapdragon® X Elite	2.393 ms	1 - 1 MB	NPU
LiteHRNet	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1.351 ms	0 - 109 MB	NPU
LiteHRNet	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	4.944 ms	1 - 81 MB	NPU
LiteHRNet	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	2.073 ms	1 - 2 MB	NPU
LiteHRNet	QNN_DLC	float	Qualcomm® SA8775P	2.656 ms	1 - 83 MB	NPU
LiteHRNet	QNN_DLC	float	Qualcomm® QCS9075	2.522 ms	1 - 3 MB	NPU
LiteHRNet	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	2.804 ms	0 - 107 MB	NPU
LiteHRNet	QNN_DLC	float	Qualcomm® SA7255P	4.944 ms	1 - 81 MB	NPU
LiteHRNet	QNN_DLC	float	Qualcomm® SA8295P	3.465 ms	0 - 83 MB	NPU
LiteHRNet	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	1.021 ms	0 - 84 MB	NPU
LiteHRNet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	2.017 ms	0 - 120 MB	NPU
LiteHRNet	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	2.699 ms	0 - 149 MB	NPU
LiteHRNet	TFLITE	float	Qualcomm® QCS8275 (Proxy)	8.804 ms	0 - 114 MB	NPU
LiteHRNet	TFLITE	float	Qualcomm® QCS8550 (Proxy)	4.22 ms	0 - 3 MB	NPU
LiteHRNet	TFLITE	float	Qualcomm® SA8775P	5.195 ms	0 - 115 MB	NPU
LiteHRNet	TFLITE	float	Qualcomm® QCS9075	5.068 ms	0 - 10 MB	NPU
LiteHRNet	TFLITE	float	Qualcomm® QCS8450 (Proxy)	5.204 ms	0 - 135 MB	NPU
LiteHRNet	TFLITE	float	Qualcomm® SA7255P	8.804 ms	0 - 114 MB	NPU
LiteHRNet	TFLITE	float	Qualcomm® SA8295P	6.322 ms	0 - 114 MB	NPU
LiteHRNet	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	2.205 ms	0 - 111 MB	NPU

License

• The license for the original implementation of LiteHRNet can be found here.

References

• Lite-HRNet: A Lightweight High-Resolution Network
• Source Model Implementation

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.