MobileNet-v2: Optimized for Qualcomm Devices

MobileNetV2 is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.

This is based on the implementation of MobileNet-v2 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
QNN_DLC	w8a16	Universal	QAIRT 2.42	Download
QNN_DLC	w8a16_mixed_int16	Universal	QAIRT 2.42	Download
QNN_DLC	w8a8	Universal	QAIRT 2.42	Download
TFLITE	float	Universal	QAIRT 2.42, TFLite 2.17.0	Download
TFLITE	w8a8	Universal	QAIRT 2.42, TFLite 2.17.0	Download

For more device-specific assets and performance metrics, visit MobileNet-v2 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 MobileNet-v2 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.image_classification

Model Stats:

• Model checkpoint: Imagenet
• Input resolution: 224x224
• Number of parameters: 3.49M
• Model size (float): 13.3 MB
• Model size (w8a16): 4.39 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
MobileNet-v2	ONNX	float	Snapdragon® X Elite	0.77 ms	7 - 7 MB	NPU
MobileNet-v2	ONNX	float	Snapdragon® 8 Gen 3 Mobile	0.523 ms	0 - 120 MB	NPU
MobileNet-v2	ONNX	float	Qualcomm® QCS8550 (Proxy)	0.801 ms	0 - 48 MB	NPU
MobileNet-v2	ONNX	float	Qualcomm® QCS9075	1.078 ms	1 - 3 MB	NPU
MobileNet-v2	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	0.465 ms	0 - 100 MB	NPU
MobileNet-v2	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.377 ms	0 - 100 MB	NPU
MobileNet-v2	QNN_DLC	float	Snapdragon® X Elite	1.048 ms	1 - 1 MB	NPU
MobileNet-v2	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	0.603 ms	0 - 51 MB	NPU
MobileNet-v2	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	2.666 ms	1 - 32 MB	NPU
MobileNet-v2	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	0.897 ms	1 - 2 MB	NPU
MobileNet-v2	QNN_DLC	float	Qualcomm® SA8775P	1.209 ms	1 - 33 MB	NPU
MobileNet-v2	QNN_DLC	float	Qualcomm® QCS9075	1.127 ms	1 - 3 MB	NPU
MobileNet-v2	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	1.715 ms	0 - 54 MB	NPU
MobileNet-v2	QNN_DLC	float	Qualcomm® SA7255P	2.666 ms	1 - 32 MB	NPU
MobileNet-v2	QNN_DLC	float	Qualcomm® SA8295P	1.468 ms	0 - 30 MB	NPU
MobileNet-v2	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	0.444 ms	1 - 31 MB	NPU
MobileNet-v2	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.333 ms	1 - 35 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Snapdragon® X Elite	0.959 ms	0 - 0 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	0.572 ms	0 - 41 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Qualcomm® QCS6490	2.276 ms	0 - 2 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	1.775 ms	0 - 29 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	0.799 ms	0 - 2 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Qualcomm® SA8775P	4.295 ms	0 - 30 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Qualcomm® QCS9075	0.982 ms	2 - 4 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Qualcomm® QCM6690	3.38 ms	0 - 141 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Qualcomm® QCS8450 (Proxy)	0.989 ms	0 - 44 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Qualcomm® SA7255P	1.775 ms	0 - 29 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Qualcomm® SA8295P	1.267 ms	0 - 27 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	0.375 ms	0 - 28 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	0.878 ms	0 - 145 MB	NPU
MobileNet-v2	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	0.301 ms	0 - 32 MB	NPU
MobileNet-v2	QNN_DLC	w8a16_mixed_int16	Snapdragon® X Elite	1.01 ms	0 - 0 MB	NPU
MobileNet-v2	QNN_DLC	w8a16_mixed_int16	Snapdragon® 8 Gen 3 Mobile	0.618 ms	0 - 41 MB	NPU
MobileNet-v2	QNN_DLC	w8a16_mixed_int16	Qualcomm® QCS8275 (Proxy)	2.007 ms	0 - 29 MB	NPU
MobileNet-v2	QNN_DLC	w8a16_mixed_int16	Qualcomm® QCS8550 (Proxy)	0.863 ms	0 - 12 MB	NPU
MobileNet-v2	QNN_DLC	w8a16_mixed_int16	Qualcomm® SA8775P	1.051 ms	0 - 31 MB	NPU
MobileNet-v2	QNN_DLC	w8a16_mixed_int16	Qualcomm® QCS9075	1.048 ms	0 - 2 MB	NPU
MobileNet-v2	QNN_DLC	w8a16_mixed_int16	Qualcomm® QCM6690	4.756 ms	0 - 142 MB	NPU
MobileNet-v2	QNN_DLC	w8a16_mixed_int16	Qualcomm® SA7255P	2.007 ms	0 - 29 MB	NPU
MobileNet-v2	QNN_DLC	w8a16_mixed_int16	Snapdragon® 8 Elite For Galaxy Mobile	0.41 ms	0 - 29 MB	NPU
MobileNet-v2	QNN_DLC	w8a16_mixed_int16	Snapdragon® 7 Gen 4 Mobile	1.004 ms	0 - 142 MB	NPU
MobileNet-v2	QNN_DLC	w8a16_mixed_int16	Snapdragon® 8 Elite Gen 5 Mobile	0.324 ms	0 - 33 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Snapdragon® X Elite	0.579 ms	0 - 0 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	0.326 ms	0 - 40 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Qualcomm® QCS6490	1.373 ms	0 - 2 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	1.062 ms	0 - 29 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	0.46 ms	0 - 1 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Qualcomm® SA8775P	0.631 ms	0 - 31 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Qualcomm® QCS9075	0.525 ms	0 - 2 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Qualcomm® QCM6690	1.76 ms	0 - 28 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	0.609 ms	0 - 41 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Qualcomm® SA7255P	1.062 ms	0 - 29 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Qualcomm® SA8295P	0.832 ms	0 - 27 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.223 ms	0 - 32 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	0.469 ms	0 - 28 MB	NPU
MobileNet-v2	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.181 ms	0 - 31 MB	NPU
MobileNet-v2	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	0.593 ms	0 - 55 MB	NPU
MobileNet-v2	TFLITE	float	Qualcomm® QCS8275 (Proxy)	2.729 ms	0 - 35 MB	NPU
MobileNet-v2	TFLITE	float	Qualcomm® QCS8550 (Proxy)	0.898 ms	0 - 2 MB	NPU
MobileNet-v2	TFLITE	float	Qualcomm® SA8775P	5.179 ms	0 - 35 MB	NPU
MobileNet-v2	TFLITE	float	Qualcomm® QCS9075	1.13 ms	0 - 10 MB	NPU
MobileNet-v2	TFLITE	float	Qualcomm® QCS8450 (Proxy)	1.708 ms	0 - 56 MB	NPU
MobileNet-v2	TFLITE	float	Qualcomm® SA7255P	2.729 ms	0 - 35 MB	NPU
MobileNet-v2	TFLITE	float	Qualcomm® SA8295P	1.519 ms	0 - 32 MB	NPU
MobileNet-v2	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	0.451 ms	0 - 40 MB	NPU
MobileNet-v2	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	0.341 ms	0 - 39 MB	NPU
MobileNet-v2	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	0.289 ms	0 - 41 MB	NPU
MobileNet-v2	TFLITE	w8a8	Qualcomm® QCS6490	1.136 ms	0 - 7 MB	NPU
MobileNet-v2	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	0.987 ms	0 - 31 MB	NPU
MobileNet-v2	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	0.408 ms	0 - 1 MB	NPU
MobileNet-v2	TFLITE	w8a8	Qualcomm® SA8775P	0.605 ms	0 - 33 MB	NPU
MobileNet-v2	TFLITE	w8a8	Qualcomm® QCS9075	0.539 ms	0 - 6 MB	NPU
MobileNet-v2	TFLITE	w8a8	Qualcomm® QCM6690	1.649 ms	0 - 27 MB	NPU
MobileNet-v2	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	0.546 ms	0 - 45 MB	NPU
MobileNet-v2	TFLITE	w8a8	Qualcomm® SA7255P	0.987 ms	0 - 31 MB	NPU
MobileNet-v2	TFLITE	w8a8	Qualcomm® SA8295P	0.808 ms	0 - 28 MB	NPU
MobileNet-v2	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.214 ms	0 - 28 MB	NPU
MobileNet-v2	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	0.421 ms	0 - 28 MB	NPU
MobileNet-v2	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.184 ms	0 - 34 MB	NPU

License

• The license for the original implementation of MobileNet-v2 can be found here.

References

• MobileNetV2: Inverted Residuals and Linear Bottlenecks
• 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.