ResNeXt101: Optimized for Qualcomm Devices

ResNeXt101 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 ResNeXt101 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 ResNeXt101 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 ResNeXt101 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: 88.7M
• Model size (float): 338 MB
• Model size (w8a8): 87.3 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
ResNeXt101	ONNX	float	Snapdragon® X Elite	6.733 ms	172 - 172 MB	NPU
ResNeXt101	ONNX	float	Snapdragon® 8 Gen 3 Mobile	4.532 ms	73 - 457 MB	NPU
ResNeXt101	ONNX	float	Qualcomm® QCS8550 (Proxy)	6.6 ms	1 - 5 MB	NPU
ResNeXt101	ONNX	float	Qualcomm® QCS9075	9.735 ms	0 - 4 MB	NPU
ResNeXt101	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	3.875 ms	0 - 175 MB	NPU
ResNeXt101	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	3.043 ms	1 - 203 MB	NPU
ResNeXt101	ONNX	float	Snapdragon® X2 Elite	3.101 ms	173 - 173 MB	NPU
ResNeXt101	ONNX	w8a8	Snapdragon® X Elite	3.13 ms	87 - 87 MB	NPU
ResNeXt101	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	2.144 ms	0 - 258 MB	NPU
ResNeXt101	ONNX	w8a8	Qualcomm® QCS6490	113.168 ms	8 - 43 MB	CPU
ResNeXt101	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	2.924 ms	0 - 100 MB	NPU
ResNeXt101	ONNX	w8a8	Qualcomm® QCS9075	3.143 ms	0 - 3 MB	NPU
ResNeXt101	ONNX	w8a8	Qualcomm® QCM6690	73.293 ms	0 - 12 MB	CPU
ResNeXt101	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	1.87 ms	0 - 219 MB	NPU
ResNeXt101	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	67.91 ms	0 - 12 MB	CPU
ResNeXt101	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	1.563 ms	0 - 215 MB	NPU
ResNeXt101	ONNX	w8a8	Snapdragon® X2 Elite	1.396 ms	87 - 87 MB	NPU
ResNeXt101	QNN_DLC	float	Snapdragon® X Elite	6.934 ms	1 - 1 MB	NPU
ResNeXt101	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	4.65 ms	0 - 368 MB	NPU
ResNeXt101	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	35.92 ms	1 - 191 MB	NPU
ResNeXt101	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	6.776 ms	1 - 3 MB	NPU
ResNeXt101	QNN_DLC	float	Qualcomm® SA8775P	45.041 ms	1 - 203 MB	NPU
ResNeXt101	QNN_DLC	float	Qualcomm® QCS9075	10.003 ms	1 - 3 MB	NPU
ResNeXt101	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	9.895 ms	0 - 313 MB	NPU
ResNeXt101	QNN_DLC	float	Qualcomm® SA7255P	35.92 ms	1 - 191 MB	NPU
ResNeXt101	QNN_DLC	float	Qualcomm® SA8295P	10.896 ms	1 - 140 MB	NPU
ResNeXt101	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	3.975 ms	0 - 189 MB	NPU
ResNeXt101	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	3.081 ms	0 - 192 MB	NPU
ResNeXt101	QNN_DLC	float	Snapdragon® X2 Elite	3.686 ms	1 - 1 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Snapdragon® X Elite	3.076 ms	0 - 0 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	2.121 ms	0 - 253 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Qualcomm® QCS6490	9.316 ms	0 - 2 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	6.52 ms	0 - 206 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	2.838 ms	0 - 2 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Qualcomm® SA8775P	3.514 ms	0 - 208 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Qualcomm® QCS9075	3.129 ms	2 - 4 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Qualcomm® QCM6690	33.196 ms	0 - 370 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	3.897 ms	0 - 252 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Qualcomm® SA7255P	6.52 ms	0 - 206 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Qualcomm® SA8295P	4.243 ms	0 - 209 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	1.817 ms	0 - 205 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	4.062 ms	0 - 239 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	1.512 ms	0 - 210 MB	NPU
ResNeXt101	QNN_DLC	w8a8	Snapdragon® X2 Elite	1.696 ms	1 - 1 MB	NPU
ResNeXt101	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	4.615 ms	0 - 571 MB	NPU
ResNeXt101	TFLITE	float	Qualcomm® QCS8275 (Proxy)	35.89 ms	0 - 396 MB	NPU
ResNeXt101	TFLITE	float	Qualcomm® QCS8550 (Proxy)	6.891 ms	0 - 3 MB	NPU
ResNeXt101	TFLITE	float	Qualcomm® SA8775P	10.283 ms	0 - 396 MB	NPU
ResNeXt101	TFLITE	float	Qualcomm® QCS9075	10.033 ms	0 - 174 MB	NPU
ResNeXt101	TFLITE	float	Qualcomm® QCS8450 (Proxy)	9.811 ms	0 - 510 MB	NPU
ResNeXt101	TFLITE	float	Qualcomm® SA7255P	35.89 ms	0 - 396 MB	NPU
ResNeXt101	TFLITE	float	Qualcomm® SA8295P	10.967 ms	0 - 336 MB	NPU
ResNeXt101	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	3.953 ms	0 - 398 MB	NPU
ResNeXt101	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	3.102 ms	0 - 401 MB	NPU
ResNeXt101	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	1.976 ms	0 - 245 MB	NPU
ResNeXt101	TFLITE	w8a8	Qualcomm® QCS6490	8.858 ms	0 - 88 MB	NPU
ResNeXt101	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	6.313 ms	0 - 210 MB	NPU
ResNeXt101	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	2.677 ms	0 - 2 MB	NPU
ResNeXt101	TFLITE	w8a8	Qualcomm® SA8775P	3.361 ms	0 - 210 MB	NPU
ResNeXt101	TFLITE	w8a8	Qualcomm® QCS9075	2.953 ms	0 - 89 MB	NPU
ResNeXt101	TFLITE	w8a8	Qualcomm® QCM6690	30.629 ms	0 - 369 MB	NPU
ResNeXt101	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	3.744 ms	0 - 250 MB	NPU
ResNeXt101	TFLITE	w8a8	Qualcomm® SA7255P	6.313 ms	0 - 210 MB	NPU
ResNeXt101	TFLITE	w8a8	Qualcomm® SA8295P	4.108 ms	0 - 212 MB	NPU
ResNeXt101	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	1.772 ms	0 - 206 MB	NPU
ResNeXt101	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	3.959 ms	0 - 225 MB	NPU
ResNeXt101	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	1.485 ms	0 - 209 MB	NPU

License

• The license for the original implementation of ResNeXt101 can be found here.

References

• Aggregated Residual Transformations for Deep Neural 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.