ResNet-3D: Optimized for Qualcomm Devices

ResNet 3D is a network with 3D convolutions used for video understanding.

This is based on the implementation of ResNet-3D 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 ResNet-3D 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-3D 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: 33.4M
• Model size (float): 127 MB
• Model size (w8a8): 32.1 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
ResNet-3D	ONNX	float	Snapdragon® X Elite	13.507 ms	63 - 63 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® 8 Gen 3 Mobile	9.671 ms	1 - 186 MB	NPU
ResNet-3D	ONNX	float	Qualcomm® QCS8550 (Proxy)	13.294 ms	0 - 80 MB	NPU
ResNet-3D	ONNX	float	Qualcomm® QCS9075	25.611 ms	2 - 7 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	7.99 ms	0 - 129 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	5.995 ms	2 - 137 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® X Elite	4.763 ms	33 - 33 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	3.409 ms	0 - 188 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS6490	269.677 ms	56 - 128 MB	CPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	4.775 ms	0 - 106 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS9075	4.921 ms	0 - 3 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCM6690	328.552 ms	42 - 50 MB	CPU
ResNet-3D	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	2.833 ms	0 - 126 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	283.764 ms	41 - 48 MB	CPU
ResNet-3D	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	2.156 ms	0 - 119 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® X Elite	13.707 ms	2 - 2 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	9.759 ms	1 - 286 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	90.869 ms	0 - 212 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	13.216 ms	2 - 4 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® SA8775P	109.977 ms	1 - 213 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS9075	26.761 ms	2 - 6 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	28.392 ms	0 - 244 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® SA7255P	90.869 ms	0 - 212 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® SA8295P	26.404 ms	0 - 193 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	7.847 ms	0 - 215 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	5.674 ms	2 - 209 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® X Elite	4.831 ms	1 - 1 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	3.35 ms	1 - 123 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS6490	19.807 ms	1 - 3 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	14.911 ms	1 - 60 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	4.528 ms	1 - 2 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® SA8775P	4.616 ms	1 - 60 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS9075	4.817 ms	1 - 3 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCM6690	91.963 ms	1 - 182 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	6.298 ms	1 - 126 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® SA7255P	14.911 ms	1 - 60 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® SA8295P	8.226 ms	1 - 64 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	2.668 ms	1 - 182 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	7.755 ms	1 - 171 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	1.893 ms	1 - 67 MB	NPU
ResNet-3D	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	222.662 ms	0 - 275 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS8275 (Proxy)	554.416 ms	0 - 218 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS8550 (Proxy)	291.578 ms	0 - 2 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® SA8775P	281.931 ms	0 - 218 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS9075	292.861 ms	0 - 69 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS8450 (Proxy)	321.139 ms	0 - 274 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® SA7255P	554.416 ms	0 - 218 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® SA8295P	350.724 ms	0 - 215 MB	NPU
ResNet-3D	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	202.517 ms	0 - 224 MB	NPU
ResNet-3D	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	184.168 ms	0 - 223 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	172.956 ms	0 - 371 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS6490	1742.956 ms	695 - 1060 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	490.909 ms	0 - 252 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	235.049 ms	0 - 2 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® SA8775P	240.05 ms	0 - 282 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS9075	208.055 ms	0 - 67 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCM6690	1483.914 ms	690 - 852 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	266.587 ms	0 - 361 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® SA7255P	490.909 ms	0 - 252 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® SA8295P	259.654 ms	0 - 254 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	142.54 ms	0 - 386 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	1173.605 ms	658 - 736 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	165.428 ms	0 - 306 MB	NPU

License

• The license for the original implementation of ResNet-3D 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.