GoogLeNet: Optimized for Qualcomm Devices

GoogLeNet 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 GoogLeNet 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 1.24.1	Download
ONNX	w8a8	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
QNN_DLC	w8a8	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.17.0	Download
TFLITE	w8a8	Universal	QAIRT 2.43, TFLite 2.17.0	Download

For more device-specific assets and performance metrics, visit GoogLeNet 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 GoogLeNet 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: 6.62M
• Model size (float): 25.3 MB
• Model size (w8a8): 6.54 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
GoogLeNet	ONNX	float	Snapdragon® X Elite	0.98 ms	13 - 13 MB	NPU
GoogLeNet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	0.568 ms	0 - 49 MB	NPU
GoogLeNet	ONNX	float	Qualcomm® QCS8550 (Proxy)	0.863 ms	1 - 3 MB	NPU
GoogLeNet	ONNX	float	Qualcomm® QCS9075	1.545 ms	1 - 3 MB	NPU
GoogLeNet	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	0.472 ms	0 - 26 MB	NPU
GoogLeNet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.396 ms	0 - 31 MB	NPU
GoogLeNet	ONNX	float	Snapdragon® X2 Elite	0.416 ms	13 - 13 MB	NPU
GoogLeNet	ONNX	w8a8	Snapdragon® X Elite	0.418 ms	7 - 7 MB	NPU
GoogLeNet	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	0.229 ms	0 - 48 MB	NPU
GoogLeNet	ONNX	w8a8	Qualcomm® QCS6490	13.814 ms	8 - 18 MB	CPU
GoogLeNet	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	0.303 ms	0 - 9 MB	NPU
GoogLeNet	ONNX	w8a8	Qualcomm® QCS9075	0.429 ms	0 - 3 MB	NPU
GoogLeNet	ONNX	w8a8	Qualcomm® QCM6690	9.195 ms	8 - 16 MB	CPU
GoogLeNet	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.193 ms	0 - 35 MB	NPU
GoogLeNet	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	7.074 ms	8 - 17 MB	CPU
GoogLeNet	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.174 ms	0 - 34 MB	NPU
GoogLeNet	ONNX	w8a8	Snapdragon® X2 Elite	0.172 ms	7 - 7 MB	NPU
GoogLeNet	QNN_DLC	float	Snapdragon® X Elite	0.956 ms	1 - 1 MB	NPU
GoogLeNet	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	0.553 ms	0 - 45 MB	NPU
GoogLeNet	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	5.002 ms	1 - 28 MB	NPU
GoogLeNet	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	0.807 ms	1 - 2 MB	NPU
GoogLeNet	QNN_DLC	float	Qualcomm® SA8775P	1.527 ms	1 - 31 MB	NPU
GoogLeNet	QNN_DLC	float	Qualcomm® QCS9075	1.516 ms	1 - 3 MB	NPU
GoogLeNet	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	1.764 ms	0 - 46 MB	NPU
GoogLeNet	QNN_DLC	float	Qualcomm® SA7255P	5.002 ms	1 - 28 MB	NPU
GoogLeNet	QNN_DLC	float	Qualcomm® SA8295P	1.779 ms	0 - 25 MB	NPU
GoogLeNet	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	0.45 ms	0 - 27 MB	NPU
GoogLeNet	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.376 ms	1 - 31 MB	NPU
GoogLeNet	QNN_DLC	float	Snapdragon® X2 Elite	0.521 ms	1 - 1 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Snapdragon® X Elite	0.343 ms	0 - 0 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	0.179 ms	0 - 42 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Qualcomm® QCS6490	1.049 ms	0 - 2 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	0.808 ms	0 - 28 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	0.241 ms	0 - 85 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Qualcomm® SA8775P	0.424 ms	0 - 29 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Qualcomm® QCS9075	0.334 ms	0 - 2 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Qualcomm® QCM6690	2.241 ms	0 - 29 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	0.42 ms	0 - 43 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Qualcomm® SA7255P	0.808 ms	0 - 28 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Qualcomm® SA8295P	0.609 ms	0 - 25 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.149 ms	0 - 27 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	0.331 ms	0 - 28 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.146 ms	0 - 30 MB	NPU
GoogLeNet	QNN_DLC	w8a8	Snapdragon® X2 Elite	0.227 ms	0 - 0 MB	NPU
GoogLeNet	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	0.55 ms	0 - 56 MB	NPU
GoogLeNet	TFLITE	float	Qualcomm® QCS8275 (Proxy)	4.964 ms	0 - 34 MB	NPU
GoogLeNet	TFLITE	float	Qualcomm® QCS8550 (Proxy)	0.809 ms	0 - 5 MB	NPU
GoogLeNet	TFLITE	float	Qualcomm® SA8775P	1.55 ms	0 - 38 MB	NPU
GoogLeNet	TFLITE	float	Qualcomm® QCS9075	1.526 ms	0 - 16 MB	NPU
GoogLeNet	TFLITE	float	Qualcomm® QCS8450 (Proxy)	1.77 ms	0 - 56 MB	NPU
GoogLeNet	TFLITE	float	Qualcomm® SA7255P	4.964 ms	0 - 34 MB	NPU
GoogLeNet	TFLITE	float	Qualcomm® SA8295P	1.791 ms	0 - 32 MB	NPU
GoogLeNet	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	0.448 ms	0 - 35 MB	NPU
GoogLeNet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	0.384 ms	0 - 38 MB	NPU
GoogLeNet	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	0.185 ms	0 - 42 MB	NPU
GoogLeNet	TFLITE	w8a8	Qualcomm® QCS6490	0.891 ms	0 - 8 MB	NPU
GoogLeNet	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	0.829 ms	0 - 26 MB	NPU
GoogLeNet	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	0.242 ms	0 - 1 MB	NPU
GoogLeNet	TFLITE	w8a8	Qualcomm® SA8775P	0.441 ms	0 - 28 MB	NPU
GoogLeNet	TFLITE	w8a8	Qualcomm® QCS9075	0.352 ms	0 - 9 MB	NPU
GoogLeNet	TFLITE	w8a8	Qualcomm® QCM6690	2.242 ms	0 - 29 MB	NPU
GoogLeNet	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	0.416 ms	0 - 43 MB	NPU
GoogLeNet	TFLITE	w8a8	Qualcomm® SA7255P	0.829 ms	0 - 26 MB	NPU
GoogLeNet	TFLITE	w8a8	Qualcomm® SA8295P	0.63 ms	0 - 25 MB	NPU
GoogLeNet	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.158 ms	0 - 31 MB	NPU
GoogLeNet	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	0.33 ms	0 - 28 MB	NPU
GoogLeNet	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.143 ms	0 - 30 MB	NPU

License

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

References

• Going Deeper with Convolutions
• 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.