Inception-v3: Optimized for Qualcomm Devices

InceptionNetV3 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 Inception-v3 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
ONNX	w8a8	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
QNN_DLC	w8a8	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.19.1	Download
TFLITE	w8a8	Universal	QAIRT 2.43, TFLite 2.19.1	Download

For more device-specific assets and performance metrics, visit Inception-v3 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 Inception-v3 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: 23.9M
• Model size (float): 90.9 MB
• Model size (w8a8): 23.3 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
Inception-v3	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.804 ms	1 - 39 MB	NPU
Inception-v3	ONNX	float	Snapdragon® X2 Elite	0.721 ms	46 - 46 MB	NPU
Inception-v3	ONNX	float	Snapdragon® X Elite	1.52 ms	46 - 46 MB	NPU
Inception-v3	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.054 ms	0 - 56 MB	NPU
Inception-v3	ONNX	float	Qualcomm® QCS8550 (Proxy)	1.49 ms	0 - 50 MB	NPU
Inception-v3	ONNX	float	Qualcomm® QCS9075	2.174 ms	0 - 4 MB	NPU
Inception-v3	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	0.928 ms	0 - 40 MB	NPU
Inception-v3	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.409 ms	0 - 57 MB	NPU
Inception-v3	ONNX	w8a8	Snapdragon® X2 Elite	0.288 ms	23 - 23 MB	NPU
Inception-v3	ONNX	w8a8	Snapdragon® X Elite	0.735 ms	23 - 23 MB	NPU
Inception-v3	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	0.51 ms	0 - 92 MB	NPU
Inception-v3	ONNX	w8a8	Qualcomm® QCS6490	24.97 ms	10 - 38 MB	CPU
Inception-v3	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	0.662 ms	0 - 74 MB	NPU
Inception-v3	ONNX	w8a8	Qualcomm® QCS9075	0.738 ms	0 - 3 MB	NPU
Inception-v3	ONNX	w8a8	Qualcomm® QCM6690	17.925 ms	12 - 21 MB	CPU
Inception-v3	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.445 ms	0 - 57 MB	NPU
Inception-v3	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	14.322 ms	17 - 26 MB	CPU
Inception-v3	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.76 ms	1 - 37 MB	NPU
Inception-v3	QNN_DLC	float	Snapdragon® X2 Elite	0.77 ms	1 - 1 MB	NPU
Inception-v3	QNN_DLC	float	Snapdragon® X Elite	1.412 ms	1 - 1 MB	NPU
Inception-v3	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1.021 ms	16 - 74 MB	NPU
Inception-v3	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	7.848 ms	1 - 33 MB	NPU
Inception-v3	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	1.372 ms	1 - 2 MB	NPU
Inception-v3	QNN_DLC	float	Qualcomm® SA8775P	2.239 ms	0 - 35 MB	NPU
Inception-v3	QNN_DLC	float	Qualcomm® QCS9075	2.053 ms	1 - 3 MB	NPU
Inception-v3	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	2.342 ms	0 - 56 MB	NPU
Inception-v3	QNN_DLC	float	Qualcomm® SA7255P	7.848 ms	1 - 33 MB	NPU
Inception-v3	QNN_DLC	float	Qualcomm® SA8295P	2.684 ms	0 - 31 MB	NPU
Inception-v3	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	0.876 ms	1 - 35 MB	NPU
Inception-v3	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.36 ms	0 - 51 MB	NPU
Inception-v3	QNN_DLC	w8a8	Snapdragon® X2 Elite	0.361 ms	0 - 0 MB	NPU
Inception-v3	QNN_DLC	w8a8	Snapdragon® X Elite	0.677 ms	0 - 0 MB	NPU
Inception-v3	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	0.464 ms	0 - 73 MB	NPU
Inception-v3	QNN_DLC	w8a8	Qualcomm® QCS6490	2.301 ms	2 - 4 MB	NPU
Inception-v3	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	0.612 ms	0 - 2 MB	NPU
Inception-v3	QNN_DLC	w8a8	Qualcomm® SA8775P	0.777 ms	0 - 51 MB	NPU
Inception-v3	QNN_DLC	w8a8	Qualcomm® QCS9075	0.626 ms	2 - 4 MB	NPU
Inception-v3	QNN_DLC	w8a8	Qualcomm® QCM6690	5.678 ms	0 - 51 MB	NPU
Inception-v3	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	0.892 ms	0 - 73 MB	NPU
Inception-v3	QNN_DLC	w8a8	Qualcomm® SA8295P	1.156 ms	0 - 45 MB	NPU
Inception-v3	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.417 ms	0 - 51 MB	NPU
Inception-v3	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	0.971 ms	0 - 50 MB	NPU
Inception-v3	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	0.76 ms	0 - 75 MB	NPU
Inception-v3	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1.025 ms	0 - 126 MB	NPU
Inception-v3	TFLITE	float	Qualcomm® QCS8275 (Proxy)	7.869 ms	0 - 73 MB	NPU
Inception-v3	TFLITE	float	Qualcomm® QCS8550 (Proxy)	1.367 ms	0 - 51 MB	NPU
Inception-v3	TFLITE	float	Qualcomm® SA8775P	2.288 ms	0 - 75 MB	NPU
Inception-v3	TFLITE	float	Qualcomm® QCS9075	2.058 ms	0 - 48 MB	NPU
Inception-v3	TFLITE	float	Qualcomm® QCS8450 (Proxy)	2.342 ms	0 - 124 MB	NPU
Inception-v3	TFLITE	float	Qualcomm® SA7255P	7.869 ms	0 - 73 MB	NPU
Inception-v3	TFLITE	float	Qualcomm® SA8295P	2.697 ms	0 - 70 MB	NPU
Inception-v3	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	0.873 ms	0 - 74 MB	NPU
Inception-v3	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.381 ms	0 - 52 MB	NPU
Inception-v3	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	0.481 ms	0 - 77 MB	NPU
Inception-v3	TFLITE	w8a8	Qualcomm® QCS6490	2.342 ms	0 - 25 MB	NPU
Inception-v3	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	1.633 ms	0 - 48 MB	NPU
Inception-v3	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	0.658 ms	0 - 7 MB	NPU
Inception-v3	TFLITE	w8a8	Qualcomm® SA8775P	0.887 ms	0 - 52 MB	NPU
Inception-v3	TFLITE	w8a8	Qualcomm® QCS9075	0.707 ms	0 - 25 MB	NPU
Inception-v3	TFLITE	w8a8	Qualcomm® QCM6690	5.823 ms	0 - 52 MB	NPU
Inception-v3	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	0.946 ms	0 - 79 MB	NPU
Inception-v3	TFLITE	w8a8	Qualcomm® SA7255P	1.633 ms	0 - 48 MB	NPU
Inception-v3	TFLITE	w8a8	Qualcomm® SA8295P	1.253 ms	0 - 47 MB	NPU
Inception-v3	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.414 ms	0 - 52 MB	NPU
Inception-v3	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	1.011 ms	0 - 51 MB	NPU

License

• The license for the original implementation of Inception-v3 can be found here.

References

• Rethinking the Inception Architecture for Computer Vision
• 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.