ResNet101: Optimized for Qualcomm Devices

ResNet101 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 ResNet101 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 ResNet101 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 ResNet101 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: 44.5M
• Model size (float): 170 MB
• Model size (w8a8): 43.9 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
ResNet101	ONNX	float	Snapdragon® X2 Elite	1.622 ms	86 - 86 MB	NPU
ResNet101	ONNX	float	Snapdragon® X Elite	3.307 ms	85 - 85 MB	NPU
ResNet101	ONNX	float	Snapdragon® 8 Gen 3 Mobile	2.238 ms	0 - 128 MB	NPU
ResNet101	ONNX	float	Qualcomm® QCS8550 (Proxy)	3.08 ms	0 - 112 MB	NPU
ResNet101	ONNX	float	Qualcomm® QCS9075	5.132 ms	0 - 4 MB	NPU
ResNet101	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	1.858 ms	0 - 75 MB	NPU
ResNet101	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1.6 ms	1 - 78 MB	NPU
ResNet101	ONNX	w8a8	Snapdragon® X2 Elite	0.566 ms	43 - 43 MB	NPU
ResNet101	ONNX	w8a8	Snapdragon® X Elite	1.294 ms	43 - 43 MB	NPU
ResNet101	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	0.983 ms	0 - 145 MB	NPU
ResNet101	ONNX	w8a8	Qualcomm® QCS6490	58.052 ms	10 - 59 MB	CPU
ResNet101	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	1.285 ms	0 - 51 MB	NPU
ResNet101	ONNX	w8a8	Qualcomm® QCS9075	1.337 ms	0 - 3 MB	NPU
ResNet101	ONNX	w8a8	Qualcomm® QCM6690	42.238 ms	10 - 20 MB	CPU
ResNet101	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.841 ms	0 - 73 MB	NPU
ResNet101	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	39.302 ms	8 - 19 MB	CPU
ResNet101	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.811 ms	0 - 76 MB	NPU
ResNet101	QNN_DLC	float	Snapdragon® X2 Elite	1.957 ms	1 - 1 MB	NPU
ResNet101	QNN_DLC	float	Snapdragon® X Elite	3.544 ms	1 - 1 MB	NPU
ResNet101	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	2.415 ms	0 - 124 MB	NPU
ResNet101	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	18.269 ms	1 - 67 MB	NPU
ResNet101	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	3.325 ms	0 - 3 MB	NPU
ResNet101	QNN_DLC	float	Qualcomm® SA8775P	5.441 ms	1 - 69 MB	NPU
ResNet101	QNN_DLC	float	Qualcomm® QCS9075	5.321 ms	3 - 5 MB	NPU
ResNet101	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	5.92 ms	0 - 92 MB	NPU
ResNet101	QNN_DLC	float	Qualcomm® SA7255P	18.269 ms	1 - 67 MB	NPU
ResNet101	QNN_DLC	float	Qualcomm® SA8295P	5.664 ms	1 - 43 MB	NPU
ResNet101	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	1.96 ms	0 - 67 MB	NPU
ResNet101	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	1.615 ms	1 - 69 MB	NPU
ResNet101	QNN_DLC	w8a8	Snapdragon® X2 Elite	0.695 ms	0 - 0 MB	NPU
ResNet101	QNN_DLC	w8a8	Snapdragon® X Elite	1.281 ms	0 - 0 MB	NPU
ResNet101	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	0.968 ms	0 - 123 MB	NPU
ResNet101	QNN_DLC	w8a8	Qualcomm® QCS6490	4.705 ms	0 - 2 MB	NPU
ResNet101	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	2.957 ms	0 - 70 MB	NPU
ResNet101	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	1.272 ms	0 - 2 MB	NPU
ResNet101	QNN_DLC	w8a8	Qualcomm® SA8775P	1.531 ms	0 - 72 MB	NPU
ResNet101	QNN_DLC	w8a8	Qualcomm® QCS9075	1.322 ms	0 - 2 MB	NPU
ResNet101	QNN_DLC	w8a8	Qualcomm® QCM6690	11.647 ms	0 - 196 MB	NPU
ResNet101	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	1.781 ms	0 - 121 MB	NPU
ResNet101	QNN_DLC	w8a8	Qualcomm® SA7255P	2.957 ms	0 - 70 MB	NPU
ResNet101	QNN_DLC	w8a8	Qualcomm® SA8295P	1.926 ms	0 - 68 MB	NPU
ResNet101	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.781 ms	0 - 68 MB	NPU
ResNet101	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	1.803 ms	0 - 80 MB	NPU
ResNet101	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.744 ms	0 - 72 MB	NPU
ResNet101	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	2.384 ms	0 - 190 MB	NPU
ResNet101	TFLITE	float	Qualcomm® QCS8275 (Proxy)	18.223 ms	0 - 124 MB	NPU
ResNet101	TFLITE	float	Qualcomm® QCS8550 (Proxy)	3.316 ms	0 - 2 MB	NPU
ResNet101	TFLITE	float	Qualcomm® SA8775P	23.558 ms	0 - 124 MB	NPU
ResNet101	TFLITE	float	Qualcomm® QCS9075	5.286 ms	0 - 88 MB	NPU
ResNet101	TFLITE	float	Qualcomm® QCS8450 (Proxy)	5.898 ms	0 - 157 MB	NPU
ResNet101	TFLITE	float	Qualcomm® SA7255P	18.223 ms	0 - 124 MB	NPU
ResNet101	TFLITE	float	Qualcomm® SA8295P	5.615 ms	0 - 97 MB	NPU
ResNet101	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	1.924 ms	0 - 129 MB	NPU
ResNet101	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1.625 ms	0 - 128 MB	NPU
ResNet101	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	0.86 ms	0 - 134 MB	NPU
ResNet101	TFLITE	w8a8	Qualcomm® QCS6490	4.23 ms	0 - 45 MB	NPU
ResNet101	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	2.733 ms	0 - 69 MB	NPU
ResNet101	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	1.145 ms	0 - 2 MB	NPU
ResNet101	TFLITE	w8a8	Qualcomm® SA8775P	1.423 ms	0 - 71 MB	NPU
ResNet101	TFLITE	w8a8	Qualcomm® QCS9075	1.17 ms	0 - 45 MB	NPU
ResNet101	TFLITE	w8a8	Qualcomm® QCM6690	11.801 ms	0 - 190 MB	NPU
ResNet101	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	1.656 ms	0 - 131 MB	NPU
ResNet101	TFLITE	w8a8	Qualcomm® SA7255P	2.733 ms	0 - 69 MB	NPU
ResNet101	TFLITE	w8a8	Qualcomm® SA8295P	1.786 ms	0 - 67 MB	NPU
ResNet101	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.736 ms	0 - 70 MB	NPU
ResNet101	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	1.66 ms	0 - 76 MB	NPU
ResNet101	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.693 ms	0 - 70 MB	NPU

License

• The license for the original implementation of ResNet101 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.