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 2.42, 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 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® 8 Elite Gen 5 Mobile	6.043 ms	2 - 200 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® X2 Elite	7.053 ms	63 - 63 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® X Elite	13.558 ms	63 - 63 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® 8 Gen 3 Mobile	9.618 ms	0 - 252 MB	NPU
ResNet-3D	ONNX	float	Qualcomm® QCS8550 (Proxy)	13.194 ms	0 - 99 MB	NPU
ResNet-3D	ONNX	float	Qualcomm® QCS9075	25.981 ms	2 - 7 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	7.889 ms	0 - 182 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	1.973 ms	0 - 174 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® X2 Elite	2.219 ms	33 - 33 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® X Elite	4.774 ms	32 - 32 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	3.323 ms	0 - 248 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS6490	269.713 ms	56 - 138 MB	CPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	4.505 ms	0 - 3 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS9075	4.705 ms	0 - 3 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCM6690	329.388 ms	30 - 36 MB	CPU
ResNet-3D	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	2.723 ms	0 - 176 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	285.412 ms	79 - 86 MB	CPU
ResNet-3D	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	5.657 ms	2 - 209 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® X2 Elite	7.08 ms	2 - 2 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® X Elite	14.072 ms	2 - 2 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	9.795 ms	0 - 283 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	91.626 ms	1 - 213 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	13.337 ms	2 - 4 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® SA8775P	110.275 ms	1 - 197 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS9075	26.881 ms	2 - 6 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	28.255 ms	0 - 245 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® SA7255P	91.626 ms	1 - 213 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® SA8295P	26.353 ms	0 - 195 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	7.833 ms	2 - 217 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	1.894 ms	0 - 64 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® X2 Elite	2.462 ms	1 - 1 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® X Elite	4.859 ms	1 - 1 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	3.387 ms	0 - 125 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS6490	19.838 ms	1 - 3 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	14.862 ms	1 - 60 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	4.516 ms	1 - 2 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® SA8775P	4.607 ms	1 - 60 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS9075	4.836 ms	1 - 3 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCM6690	92.315 ms	1 - 184 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	5.864 ms	1 - 126 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® SA7255P	14.862 ms	1 - 60 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® SA8295P	8.274 ms	1 - 62 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	2.657 ms	0 - 179 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	7.782 ms	1 - 171 MB	NPU
ResNet-3D	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	185.721 ms	0 - 220 MB	NPU
ResNet-3D	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	206.352 ms	0 - 280 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS8275 (Proxy)	555.868 ms	0 - 219 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS8550 (Proxy)	312.04 ms	0 - 3 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® SA8775P	277.909 ms	0 - 220 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS9075	297.687 ms	0 - 69 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS8450 (Proxy)	301.954 ms	0 - 276 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® SA7255P	555.868 ms	0 - 219 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® SA8295P	336.411 ms	0 - 214 MB	NPU
ResNet-3D	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	204.723 ms	0 - 223 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	168.686 ms	0 - 301 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	179.08 ms	0 - 323 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	493.289 ms	0 - 252 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	242.365 ms	0 - 3 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® SA8775P	246.015 ms	0 - 280 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS9075	208.745 ms	0 - 67 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCM6690	1452.863 ms	659 - 813 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	267.43 ms	0 - 358 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® SA7255P	493.289 ms	0 - 252 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® SA8295P	256.046 ms	0 - 253 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	136.261 ms	0 - 258 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	1175.915 ms	660 - 721 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.