MediaPipe-Hand-Detection: Optimized for Qualcomm Devices

The MediaPipe Hand Landmark Detector is a machine learning pipeline that predicts bounding boxes and pose skeletons of hands in an image.

This is based on the implementation of MediaPipe-Hand-Detection 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 2.37, ONNX Runtime 1.23.0	Download
QNN_DLC	float	Universal	QAIRT 2.42	Download
TFLITE	float	Universal	QAIRT 2.42, TFLite 2.17.0	Download

For more device-specific assets and performance metrics, visit MediaPipe-Hand-Detection 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 MediaPipe-Hand-Detection on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.object_detection

Model Stats:

• Input resolution: 256x256
• Number of parameters (HandDetector): 1.76M
• Model size (HandDetector) (float): 6.75 MB
• Number of parameters (HandLandmarkDetector): 2.01M
• Model size (HandLandmarkDetector) (float): 7.70 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
HandDetector	ONNX	float	Snapdragon® X Elite	0.97 ms	3 - 3 MB	NPU
HandDetector	ONNX	float	Snapdragon® 8 Gen 3 Mobile	0.823 ms	0 - 134 MB	NPU
HandDetector	ONNX	float	Qualcomm® QCS8550 (Proxy)	1.047 ms	0 - 52 MB	NPU
HandDetector	ONNX	float	Qualcomm® QCS9075	1.474 ms	1 - 4 MB	NPU
HandDetector	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	0.632 ms	0 - 110 MB	NPU
HandDetector	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.588 ms	0 - 110 MB	NPU
HandDetector	QNN_DLC	float	Snapdragon® X Elite	0.895 ms	1 - 1 MB	NPU
HandDetector	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	0.53 ms	0 - 60 MB	NPU
HandDetector	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	3.785 ms	1 - 41 MB	NPU
HandDetector	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	0.731 ms	1 - 3 MB	NPU
HandDetector	QNN_DLC	float	Qualcomm® SA8775P	1.269 ms	1 - 43 MB	NPU
HandDetector	QNN_DLC	float	Qualcomm® QCS9075	1.123 ms	1 - 3 MB	NPU
HandDetector	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	1.349 ms	0 - 54 MB	NPU
HandDetector	QNN_DLC	float	Qualcomm® SA7255P	3.785 ms	1 - 41 MB	NPU
HandDetector	QNN_DLC	float	Qualcomm® SA8295P	1.71 ms	0 - 31 MB	NPU
HandDetector	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	0.442 ms	1 - 40 MB	NPU
HandDetector	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.381 ms	1 - 43 MB	NPU
HandDetector	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	0.531 ms	0 - 57 MB	NPU
HandDetector	TFLITE	float	Qualcomm® QCS8275 (Proxy)	3.808 ms	0 - 39 MB	NPU
HandDetector	TFLITE	float	Qualcomm® QCS8550 (Proxy)	0.73 ms	0 - 2 MB	NPU
HandDetector	TFLITE	float	Qualcomm® SA8775P	1.302 ms	0 - 43 MB	NPU
HandDetector	TFLITE	float	Qualcomm® QCS9075	1.124 ms	0 - 7 MB	NPU
HandDetector	TFLITE	float	Qualcomm® QCS8450 (Proxy)	1.401 ms	0 - 51 MB	NPU
HandDetector	TFLITE	float	Qualcomm® SA7255P	3.808 ms	0 - 39 MB	NPU
HandDetector	TFLITE	float	Qualcomm® SA8295P	1.702 ms	0 - 30 MB	NPU
HandDetector	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	0.449 ms	0 - 44 MB	NPU
HandDetector	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	0.381 ms	0 - 43 MB	NPU
HandLandmarkDetector	ONNX	float	Snapdragon® X Elite	1.349 ms	6 - 6 MB	NPU
HandLandmarkDetector	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.058 ms	0 - 125 MB	NPU
HandLandmarkDetector	ONNX	float	Qualcomm® QCS8550 (Proxy)	1.39 ms	0 - 13 MB	NPU
HandLandmarkDetector	ONNX	float	Qualcomm® QCS9075	2.125 ms	1 - 4 MB	NPU
HandLandmarkDetector	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	0.806 ms	0 - 106 MB	NPU
HandLandmarkDetector	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.716 ms	0 - 106 MB	NPU
HandLandmarkDetector	QNN_DLC	float	Snapdragon® X Elite	1.265 ms	1 - 1 MB	NPU
HandLandmarkDetector	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	0.763 ms	0 - 58 MB	NPU
HandLandmarkDetector	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	5.323 ms	1 - 35 MB	NPU
HandLandmarkDetector	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	1.009 ms	1 - 3 MB	NPU
HandLandmarkDetector	QNN_DLC	float	Qualcomm® SA8775P	1.851 ms	1 - 39 MB	NPU
HandLandmarkDetector	QNN_DLC	float	Qualcomm® QCS9075	1.677 ms	3 - 5 MB	NPU
HandLandmarkDetector	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	1.942 ms	0 - 53 MB	NPU
HandLandmarkDetector	QNN_DLC	float	Qualcomm® SA7255P	5.323 ms	1 - 35 MB	NPU
HandLandmarkDetector	QNN_DLC	float	Qualcomm® SA8295P	2.261 ms	0 - 31 MB	NPU
HandLandmarkDetector	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	0.607 ms	0 - 38 MB	NPU
HandLandmarkDetector	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.512 ms	1 - 39 MB	NPU
HandLandmarkDetector	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	0.754 ms	0 - 61 MB	NPU
HandLandmarkDetector	TFLITE	float	Qualcomm® QCS8275 (Proxy)	5.363 ms	0 - 42 MB	NPU
HandLandmarkDetector	TFLITE	float	Qualcomm® QCS8550 (Proxy)	0.996 ms	0 - 2 MB	NPU
HandLandmarkDetector	TFLITE	float	Qualcomm® SA8775P	1.882 ms	0 - 45 MB	NPU
HandLandmarkDetector	TFLITE	float	Qualcomm® QCS9075	1.682 ms	0 - 9 MB	NPU
HandLandmarkDetector	TFLITE	float	Qualcomm® QCS8450 (Proxy)	1.938 ms	0 - 57 MB	NPU
HandLandmarkDetector	TFLITE	float	Qualcomm® SA7255P	5.363 ms	0 - 42 MB	NPU
HandLandmarkDetector	TFLITE	float	Qualcomm® SA8295P	2.273 ms	0 - 34 MB	NPU
HandLandmarkDetector	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	0.604 ms	0 - 44 MB	NPU
HandLandmarkDetector	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	0.508 ms	0 - 45 MB	NPU

License

• The license for the original implementation of MediaPipe-Hand-Detection can be found here.

References

• MediaPipe Hands: On-device Real-time Hand Tracking
• 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.