ResNet-Mixed-Convolution: Optimized for Qualcomm Devices

ResNet Mixed Convolutions is a network with a mixture of 2D and 3D convolutions used for video understanding.

This is based on the implementation of ResNet-Mixed-Convolution 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	w8a16	Universal	QAIRT 2.37, ONNX Runtime 1.23.0	Download
QNN_DLC	float	Universal	QAIRT 2.42	Download
QNN_DLC	w8a16	Universal	QAIRT 2.42	Download
TFLITE	float	Universal	QAIRT 2.42, TFLite 2.17.0	Download

For more device-specific assets and performance metrics, visit ResNet-Mixed-Convolution 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 ResNet-Mixed-Convolution on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.video_classification

Model Stats:

• Model checkpoint: Kinetics-400
• Input resolution: 112x112
• Number of parameters: 11.7M
• Model size (float): 44.6 MB
• Model size (w8a16): 11.5 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
ResNet-Mixed-Convolution	ONNX	float	Snapdragon® X Elite	14.062 ms	22 - 22 MB	NPU
ResNet-Mixed-Convolution	ONNX	float	Snapdragon® 8 Gen 3 Mobile	9.988 ms	2 - 235 MB	NPU
ResNet-Mixed-Convolution	ONNX	float	Qualcomm® QCS8550 (Proxy)	13.721 ms	0 - 38 MB	NPU
ResNet-Mixed-Convolution	ONNX	float	Qualcomm® QCS9075	27.604 ms	2 - 5 MB	NPU
ResNet-Mixed-Convolution	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	8.173 ms	0 - 167 MB	NPU
ResNet-Mixed-Convolution	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	5.86 ms	2 - 167 MB	NPU
ResNet-Mixed-Convolution	ONNX	w8a16	Snapdragon® X Elite	9.123 ms	12 - 12 MB	NPU
ResNet-Mixed-Convolution	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	6.522 ms	1 - 195 MB	NPU
ResNet-Mixed-Convolution	ONNX	w8a16	Qualcomm® QCS6490	1728.182 ms	45 - 62 MB	CPU
ResNet-Mixed-Convolution	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	8.78 ms	0 - 232 MB	NPU
ResNet-Mixed-Convolution	ONNX	w8a16	Qualcomm® QCS9075	9.303 ms	1 - 4 MB	NPU
ResNet-Mixed-Convolution	ONNX	w8a16	Qualcomm® QCM6690	890.317 ms	106 - 113 MB	CPU
ResNet-Mixed-Convolution	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	5.341 ms	1 - 135 MB	NPU
ResNet-Mixed-Convolution	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	881.492 ms	105 - 112 MB	CPU
ResNet-Mixed-Convolution	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	3.904 ms	0 - 138 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	float	Snapdragon® X Elite	14.152 ms	2 - 2 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	9.536 ms	0 - 282 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	13.367 ms	2 - 5 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	float	Qualcomm® SA8775P	25.055 ms	1 - 223 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	float	Qualcomm® QCS9075	27.978 ms	2 - 6 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	27.577 ms	0 - 240 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	float	Qualcomm® SA8295P	26.689 ms	0 - 192 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	7.684 ms	2 - 227 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	5.76 ms	2 - 233 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Snapdragon® X Elite	9.877 ms	1 - 1 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	6.654 ms	1 - 255 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Qualcomm® QCS6490	37.276 ms	1 - 4 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	9.22 ms	1 - 3 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Qualcomm® SA8775P	9.288 ms	1 - 191 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Qualcomm® QCS9075	10.377 ms	1 - 4 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Qualcomm® QCM6690	178.522 ms	1 - 208 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Qualcomm® QCS8450 (Proxy)	12.996 ms	1 - 255 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Qualcomm® SA8295P	16.239 ms	1 - 192 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	5.261 ms	1 - 186 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	15.94 ms	1 - 197 MB	NPU
ResNet-Mixed-Convolution	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	3.995 ms	1 - 190 MB	NPU
ResNet-Mixed-Convolution	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	218.735 ms	0 - 313 MB	NPU
ResNet-Mixed-Convolution	TFLITE	float	Qualcomm® QCS8275 (Proxy)	585.16 ms	0 - 252 MB	NPU
ResNet-Mixed-Convolution	TFLITE	float	Qualcomm® QCS8550 (Proxy)	313.312 ms	0 - 2 MB	NPU
ResNet-Mixed-Convolution	TFLITE	float	Qualcomm® SA8775P	296.864 ms	0 - 252 MB	NPU
ResNet-Mixed-Convolution	TFLITE	float	Qualcomm® QCS9075	321.486 ms	0 - 28 MB	NPU
ResNet-Mixed-Convolution	TFLITE	float	Qualcomm® QCS8450 (Proxy)	345.498 ms	0 - 277 MB	NPU
ResNet-Mixed-Convolution	TFLITE	float	Qualcomm® SA7255P	585.16 ms	0 - 252 MB	NPU
ResNet-Mixed-Convolution	TFLITE	float	Qualcomm® SA8295P	353.223 ms	0 - 234 MB	NPU
ResNet-Mixed-Convolution	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	203.459 ms	0 - 257 MB	NPU
ResNet-Mixed-Convolution	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	189.334 ms	0 - 254 MB	NPU

License

• The license for the original implementation of ResNet-Mixed-Convolution can be found here.

References

• A Closer Look at Spatiotemporal Convolutions for Action 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.