WideResNet50: Optimized for Qualcomm Devices

WideResNet50 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 WideResNet50 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
ONNX	w8a8	Universal	QAIRT 2.37, ONNX Runtime 1.23.0	Download
QNN_DLC	float	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 WideResNet50 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 WideResNet50 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: 68.9M
• Model size (float): 263 MB
• Model size (w8a8): 66.6 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
WideResNet50	ONNX	float	Snapdragon® X Elite	4.484 ms	133 - 133 MB	NPU
WideResNet50	ONNX	float	Snapdragon® 8 Gen 3 Mobile	3.589 ms	0 - 167 MB	NPU
WideResNet50	ONNX	float	Qualcomm® QCS8550 (Proxy)	4.819 ms	1 - 3 MB	NPU
WideResNet50	ONNX	float	Qualcomm® QCS9075	7.314 ms	0 - 4 MB	NPU
WideResNet50	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	3.031 ms	0 - 118 MB	NPU
WideResNet50	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	2.788 ms	1 - 127 MB	NPU
WideResNet50	ONNX	w8a8	Snapdragon® X Elite	1.784 ms	67 - 67 MB	NPU
WideResNet50	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	1.529 ms	0 - 195 MB	NPU
WideResNet50	ONNX	w8a8	Qualcomm® QCS6490	79.763 ms	10 - 119 MB	CPU
WideResNet50	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	2.058 ms	0 - 171 MB	NPU
WideResNet50	ONNX	w8a8	Qualcomm® QCS9075	2.124 ms	0 - 3 MB	NPU
WideResNet50	ONNX	w8a8	Qualcomm® QCM6690	61.366 ms	8 - 17 MB	CPU
WideResNet50	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	1.404 ms	0 - 121 MB	NPU
WideResNet50	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	58.11 ms	10 - 19 MB	CPU
WideResNet50	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	1.266 ms	0 - 122 MB	NPU
WideResNet50	QNN_DLC	float	Snapdragon® X Elite	4.693 ms	1 - 1 MB	NPU
WideResNet50	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	3.627 ms	1 - 224 MB	NPU
WideResNet50	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	24.44 ms	1 - 173 MB	NPU
WideResNet50	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	4.761 ms	1 - 3 MB	NPU
WideResNet50	QNN_DLC	float	Qualcomm® SA8775P	7.123 ms	1 - 173 MB	NPU
WideResNet50	QNN_DLC	float	Qualcomm® QCS9075	7.067 ms	1 - 3 MB	NPU
WideResNet50	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	9.191 ms	0 - 197 MB	NPU
WideResNet50	QNN_DLC	float	Qualcomm® SA7255P	24.44 ms	1 - 173 MB	NPU
WideResNet50	QNN_DLC	float	Qualcomm® SA8295P	7.86 ms	1 - 152 MB	NPU
WideResNet50	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	2.968 ms	1 - 175 MB	NPU
WideResNet50	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	2.497 ms	1 - 171 MB	NPU
WideResNet50	QNN_DLC	w8a8	Snapdragon® X Elite	1.797 ms	0 - 0 MB	NPU
WideResNet50	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	1.432 ms	0 - 127 MB	NPU
WideResNet50	QNN_DLC	w8a8	Qualcomm® QCS6490	7.754 ms	2 - 4 MB	NPU
WideResNet50	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	1.875 ms	0 - 2 MB	NPU
WideResNet50	QNN_DLC	w8a8	Qualcomm® SA8775P	1.992 ms	0 - 53 MB	NPU
WideResNet50	QNN_DLC	w8a8	Qualcomm® QCS9075	1.907 ms	2 - 4 MB	NPU
WideResNet50	QNN_DLC	w8a8	Qualcomm® QCM6690	20.895 ms	0 - 212 MB	NPU
WideResNet50	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	2.537 ms	0 - 127 MB	NPU
WideResNet50	QNN_DLC	w8a8	Qualcomm® SA8295P	2.762 ms	0 - 53 MB	NPU
WideResNet50	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	1.199 ms	0 - 50 MB	NPU
WideResNet50	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	2.902 ms	0 - 209 MB	NPU
WideResNet50	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	1.091 ms	0 - 53 MB	NPU
WideResNet50	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	3.646 ms	0 - 258 MB	NPU
WideResNet50	TFLITE	float	Qualcomm® QCS8275 (Proxy)	23.88 ms	0 - 117 MB	NPU
WideResNet50	TFLITE	float	Qualcomm® QCS8550 (Proxy)	4.824 ms	0 - 2 MB	NPU
WideResNet50	TFLITE	float	Qualcomm® SA8775P	7.081 ms	0 - 117 MB	NPU
WideResNet50	TFLITE	float	Qualcomm® QCS9075	7.049 ms	0 - 134 MB	NPU
WideResNet50	TFLITE	float	Qualcomm® QCS8450 (Proxy)	8.173 ms	0 - 228 MB	NPU
WideResNet50	TFLITE	float	Qualcomm® SA7255P	23.88 ms	0 - 117 MB	NPU
WideResNet50	TFLITE	float	Qualcomm® SA8295P	7.773 ms	0 - 101 MB	NPU
WideResNet50	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	2.877 ms	0 - 121 MB	NPU
WideResNet50	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	2.618 ms	0 - 119 MB	NPU
WideResNet50	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	1.318 ms	0 - 128 MB	NPU
WideResNet50	TFLITE	w8a8	Qualcomm® QCS6490	6.782 ms	0 - 68 MB	NPU
WideResNet50	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	3.749 ms	0 - 50 MB	NPU
WideResNet50	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	1.712 ms	0 - 3 MB	NPU
WideResNet50	TFLITE	w8a8	Qualcomm® SA8775P	1.89 ms	0 - 52 MB	NPU
WideResNet50	TFLITE	w8a8	Qualcomm® QCS9075	1.735 ms	0 - 68 MB	NPU
WideResNet50	TFLITE	w8a8	Qualcomm® QCM6690	20.013 ms	0 - 211 MB	NPU
WideResNet50	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	2.398 ms	0 - 129 MB	NPU
WideResNet50	TFLITE	w8a8	Qualcomm® SA7255P	3.749 ms	0 - 50 MB	NPU
WideResNet50	TFLITE	w8a8	Qualcomm® SA8295P	2.627 ms	0 - 48 MB	NPU
WideResNet50	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	1.122 ms	0 - 52 MB	NPU
WideResNet50	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	2.735 ms	0 - 205 MB	NPU
WideResNet50	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	1.049 ms	0 - 53 MB	NPU

License

• The license for the original implementation of WideResNet50 can be found here.

References

• Wide Residual Networks
• 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.