ResNet50: Optimized for Qualcomm Devices

ResNet50 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 ResNet50 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 ResNet50 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 ResNet50 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: 25.5M
• Model size (float): 97.4 MB
• Model size (w8a8): 25.1 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
ResNet50	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1.063 ms	1 - 75 MB	NPU
ResNet50	ONNX	float	Snapdragon® X2 Elite	0.989 ms	49 - 49 MB	NPU
ResNet50	ONNX	float	Snapdragon® X Elite	2.091 ms	49 - 49 MB	NPU
ResNet50	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.476 ms	0 - 79 MB	NPU
ResNet50	ONNX	float	Qualcomm® QCS8550 (Proxy)	1.994 ms	1 - 3 MB	NPU
ResNet50	ONNX	float	Qualcomm® QCS9075	3.163 ms	0 - 4 MB	NPU
ResNet50	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	1.247 ms	0 - 44 MB	NPU
ResNet50	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.557 ms	0 - 45 MB	NPU
ResNet50	ONNX	w8a8	Snapdragon® X2 Elite	0.411 ms	25 - 25 MB	NPU
ResNet50	ONNX	w8a8	Snapdragon® X Elite	0.964 ms	25 - 25 MB	NPU
ResNet50	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	0.674 ms	0 - 83 MB	NPU
ResNet50	ONNX	w8a8	Qualcomm® QCS6490	31.435 ms	8 - 27 MB	CPU
ResNet50	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	0.927 ms	0 - 35 MB	NPU
ResNet50	ONNX	w8a8	Qualcomm® QCS9075	0.973 ms	0 - 3 MB	NPU
ResNet50	ONNX	w8a8	Qualcomm® QCM6690	23.518 ms	4 - 12 MB	CPU
ResNet50	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.585 ms	0 - 44 MB	NPU
ResNet50	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	17.843 ms	7 - 15 MB	CPU
ResNet50	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	1.083 ms	1 - 51 MB	NPU
ResNet50	QNN_DLC	float	Snapdragon® X2 Elite	1.244 ms	1 - 1 MB	NPU
ResNet50	QNN_DLC	float	Snapdragon® X Elite	2.305 ms	1 - 1 MB	NPU
ResNet50	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1.616 ms	1 - 84 MB	NPU
ResNet50	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	10.638 ms	1 - 46 MB	NPU
ResNet50	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	2.19 ms	1 - 2 MB	NPU
ResNet50	QNN_DLC	float	Qualcomm® SA8775P	3.362 ms	1 - 48 MB	NPU
ResNet50	QNN_DLC	float	Qualcomm® QCS9075	3.319 ms	1 - 3 MB	NPU
ResNet50	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	3.749 ms	0 - 62 MB	NPU
ResNet50	QNN_DLC	float	Qualcomm® SA7255P	10.638 ms	1 - 46 MB	NPU
ResNet50	QNN_DLC	float	Qualcomm® SA8295P	3.668 ms	0 - 30 MB	NPU
ResNet50	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	1.277 ms	0 - 50 MB	NPU
ResNet50	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.481 ms	0 - 41 MB	NPU
ResNet50	QNN_DLC	w8a8	Snapdragon® X2 Elite	0.472 ms	0 - 0 MB	NPU
ResNet50	QNN_DLC	w8a8	Snapdragon® X Elite	0.953 ms	0 - 0 MB	NPU
ResNet50	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	0.675 ms	0 - 70 MB	NPU
ResNet50	QNN_DLC	w8a8	Qualcomm® QCS6490	3.007 ms	0 - 2 MB	NPU
ResNet50	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	1.984 ms	0 - 41 MB	NPU
ResNet50	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	0.9 ms	0 - 2 MB	NPU
ResNet50	QNN_DLC	w8a8	Qualcomm® SA8775P	4.143 ms	0 - 41 MB	NPU
ResNet50	QNN_DLC	w8a8	Qualcomm® QCS9075	0.971 ms	2 - 4 MB	NPU
ResNet50	QNN_DLC	w8a8	Qualcomm® QCM6690	6.483 ms	0 - 161 MB	NPU
ResNet50	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	1.195 ms	0 - 72 MB	NPU
ResNet50	QNN_DLC	w8a8	Qualcomm® SA7255P	1.984 ms	0 - 41 MB	NPU
ResNet50	QNN_DLC	w8a8	Qualcomm® SA8295P	1.42 ms	0 - 39 MB	NPU
ResNet50	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.526 ms	0 - 38 MB	NPU
ResNet50	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	1.235 ms	0 - 48 MB	NPU
ResNet50	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1.083 ms	0 - 83 MB	NPU
ResNet50	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1.581 ms	0 - 120 MB	NPU
ResNet50	TFLITE	float	Qualcomm® QCS8275 (Proxy)	10.508 ms	0 - 80 MB	NPU
ResNet50	TFLITE	float	Qualcomm® QCS8550 (Proxy)	2.201 ms	0 - 2 MB	NPU
ResNet50	TFLITE	float	Qualcomm® SA8775P	3.358 ms	0 - 81 MB	NPU
ResNet50	TFLITE	float	Qualcomm® QCS9075	3.352 ms	0 - 52 MB	NPU
ResNet50	TFLITE	float	Qualcomm® QCS8450 (Proxy)	3.731 ms	0 - 108 MB	NPU
ResNet50	TFLITE	float	Qualcomm® SA7255P	10.508 ms	0 - 80 MB	NPU
ResNet50	TFLITE	float	Qualcomm® SA8295P	3.61 ms	0 - 62 MB	NPU
ResNet50	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	1.26 ms	0 - 82 MB	NPU
ResNet50	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.439 ms	0 - 41 MB	NPU
ResNet50	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	0.581 ms	0 - 73 MB	NPU
ResNet50	TFLITE	w8a8	Qualcomm® QCS6490	2.6 ms	0 - 27 MB	NPU
ResNet50	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	1.772 ms	0 - 40 MB	NPU
ResNet50	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	0.764 ms	0 - 8 MB	NPU
ResNet50	TFLITE	w8a8	Qualcomm® SA8775P	0.983 ms	0 - 42 MB	NPU
ResNet50	TFLITE	w8a8	Qualcomm® QCS9075	0.818 ms	0 - 27 MB	NPU
ResNet50	TFLITE	w8a8	Qualcomm® QCM6690	6.147 ms	0 - 160 MB	NPU
ResNet50	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	1.06 ms	0 - 73 MB	NPU
ResNet50	TFLITE	w8a8	Qualcomm® SA7255P	1.772 ms	0 - 40 MB	NPU
ResNet50	TFLITE	w8a8	Qualcomm® SA8295P	1.302 ms	0 - 37 MB	NPU
ResNet50	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.482 ms	0 - 42 MB	NPU
ResNet50	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	1.088 ms	0 - 47 MB	NPU

License

• The license for the original implementation of ResNet50 can be found here.

References

• Deep Residual Learning for Image Recognition
• 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.