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 2.45, ONNX Runtime 1.25.0	Download
ONNX	w8a8	Universal	QAIRT 2.45, ONNX Runtime 1.25.0	Download
QNN_DLC	float	Universal	QAIRT 2.45	Download
QNN_DLC	w8a8	Universal	QAIRT 2.45	Download
TFLITE	float	Universal	QAIRT 2.45	Download
TFLITE	w8a8	Universal	QAIRT 2.45	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® X2 Elite	35.767 ms	201 - 201 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® X Elite	67.533 ms	171 - 171 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® 8 Gen 3 Mobile	50.354 ms	12 - 584 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® 8 Gen 1 Mobile	134.238 ms	12 - 490 MB	NPU
ResNet-3D	ONNX	float	Qualcomm® QCS8550 (Proxy)	67.28 ms	12 - 16 MB	NPU
ResNet-3D	ONNX	float	Qualcomm® QCS8450	134.238 ms	12 - 490 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® 8 Elite Mobile	41.575 ms	1 - 388 MB	NPU
ResNet-3D	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	29.505 ms	1 - 403 MB	NPU
ResNet-3D	ONNX	float	Qualcomm® QCS9075	128.384 ms	11 - 68 MB	NPU
ResNet-3D	ONNX	float	Qualcomm® QCS8750	41.575 ms	1 - 388 MB	NPU
ResNet-3D	ONNX	float	Qualcomm® QCS7181	67.533 ms	171 - 171 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® X2 Elite	11.525 ms	210 - 210 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® X Elite	22.567 ms	180 - 180 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	16.827 ms	3 - 406 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® 8 Gen 1 Mobile	37.595 ms	3 - 410 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	22.277 ms	3 - 7 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS8450	37.595 ms	3 - 410 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	9.977 ms	0 - 293 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	42.401 ms	3 - 291 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS9075	22.06 ms	3 - 48 MB	NPU
ResNet-3D	ONNX	w8a8	Snapdragon® 8 Elite Mobile	15.002 ms	0 - 291 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS7790	42.401 ms	3 - 291 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS8750	15.002 ms	0 - 291 MB	NPU
ResNet-3D	ONNX	w8a8	Qualcomm® QCS7181	22.567 ms	180 - 180 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® X2 Elite	36.369 ms	12 - 12 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® X Elite	70.634 ms	12 - 12 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	51.477 ms	12 - 602 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	134.622 ms	7 - 500 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS8275	467.796 ms	1 - 395 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	69.052 ms	12 - 15 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS8450	134.622 ms	7 - 500 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® 8 Elite Mobile	41.726 ms	11 - 400 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® SA8295P	133.012 ms	0 - 317 MB	NPU
ResNet-3D	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	29.444 ms	12 - 406 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® SA7255P	467.796 ms	1 - 395 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS9075	130.694 ms	12 - 25 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS8750	41.726 ms	11 - 400 MB	NPU
ResNet-3D	QNN_DLC	float	Qualcomm® QCS7181	70.634 ms	12 - 12 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® X2 Elite	12.349 ms	3 - 3 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® X Elite	23.392 ms	3 - 3 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	17.043 ms	3 - 409 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 8 Gen 1 Mobile	36.78 ms	3 - 412 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS6490	102.317 ms	2 - 7 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS8275	75.894 ms	3 - 284 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	21.484 ms	3 - 5 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS8450	36.78 ms	3 - 412 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	10.65 ms	3 - 298 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	43.178 ms	3 - 302 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS9075	22.559 ms	5 - 9 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® SA7255P	75.894 ms	3 - 284 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCM6690	450.668 ms	46 - 402 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Snapdragon® 8 Elite Mobile	15.271 ms	3 - 291 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® SA8295P	39.848 ms	0 - 281 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS7790	43.178 ms	3 - 302 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS8750	15.271 ms	3 - 291 MB	NPU
ResNet-3D	QNN_DLC	w8a8	Qualcomm® QCS7181	23.392 ms	3 - 3 MB	NPU
ResNet-3D	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1144.915 ms	0 - 640 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS8275	2783.244 ms	1 - 456 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS8550 (Proxy)	1429.804 ms	0 - 3 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® SA8775P	188084.492 ms	910 - 926 MB	CPU
ResNet-3D	TFLITE	float	Qualcomm® SA8650P	188084.492 ms	910 - 926 MB	CPU
ResNet-3D	TFLITE	float	Qualcomm® SA8255P	188084.492 ms	910 - 926 MB	CPU
ResNet-3D	TFLITE	float	Snapdragon® 8 Elite Mobile	1009.106 ms	1 - 467 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® SA8295P	1728.57 ms	1 - 456 MB	NPU
ResNet-3D	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	887.207 ms	0 - 471 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® SA7255P	2783.244 ms	1 - 456 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS9075	1457.644 ms	0 - 82 MB	NPU
ResNet-3D	TFLITE	float	Qualcomm® QCS8750	1009.106 ms	1 - 467 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	950.777 ms	0 - 1007 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 8 Gen 1 Mobile	1414.799 ms	1 - 869 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS6490	9566.046 ms	3462 - 5277 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS8275	2556.985 ms	1 - 819 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	1255.964 ms	0 - 3 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® SA8775P	191859.909 ms	918 - 934 MB	CPU
ResNet-3D	TFLITE	w8a8	Qualcomm® SA8650P	191859.909 ms	918 - 934 MB	CPU
ResNet-3D	TFLITE	w8a8	Qualcomm® SA8255P	191859.909 ms	918 - 934 MB	CPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS8450	1414.799 ms	1 - 869 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	817.083 ms	0 - 928 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	343740.245 ms	919 - 1201 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS9075	1116.476 ms	0 - 74 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® SA7255P	2556.985 ms	1 - 819 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCM6690	386810.527 ms	905 - 1116 MB	NPU
ResNet-3D	TFLITE	w8a8	Snapdragon® 8 Elite Mobile	656.895 ms	0 - 729 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® SA8295P	1342.31 ms	0 - 685 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS7790	343740.245 ms	919 - 1201 MB	NPU
ResNet-3D	TFLITE	w8a8	Qualcomm® QCS8750	656.895 ms	0 - 729 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.