ResNet-2Plus1D: Optimized for Qualcomm Devices

ResNet (2+1)D Convolutions is a network which explicitly factorizes 3D convolution into two separate and successive operations, a 2D spatial convolution and a 1D temporal convolution. It used for video understanding applications.

This is based on the implementation of ResNet-2Plus1D 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 1.24.3	Download
ONNX	w8a8	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
QNN_DLC	w8a8	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.19.1	Download
TFLITE	w8a8	Universal	QAIRT 2.43, TFLite 2.19.1	Download

For more device-specific assets and performance metrics, visit ResNet-2Plus1D 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-2Plus1D 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: 31.5M
• Model size (float): 120 MB
• Model size (w8a8): 30.8 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	5.515 ms	2 - 215 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® X2 Elite	6.147 ms	60 - 60 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® X Elite	12.211 ms	60 - 60 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Gen 3 Mobile	9.055 ms	1 - 292 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® QCS8550 (Proxy)	12.005 ms	2 - 5 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® QCS9075	21.356 ms	2 - 7 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	7.225 ms	0 - 211 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	1.926 ms	0 - 191 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® X2 Elite	2.041 ms	31 - 31 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® X Elite	4.525 ms	31 - 31 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	3.291 ms	0 - 226 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS6490	325.252 ms	97 - 127 MB	CPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	4.289 ms	0 - 35 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS9075	4.105 ms	1 - 3 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCM6690	301.71 ms	100 - 108 MB	CPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	2.636 ms	0 - 190 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	266.883 ms	100 - 108 MB	CPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	5.707 ms	2 - 228 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X2 Elite	6.64 ms	2 - 2 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X Elite	13.081 ms	2 - 2 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	9.202 ms	0 - 302 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	81.818 ms	1 - 215 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	12.749 ms	2 - 297 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8775P	21.319 ms	1 - 216 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS9075	22.984 ms	4 - 8 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	28.825 ms	0 - 278 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA7255P	81.818 ms	1 - 215 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8295P	22.732 ms	0 - 198 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	7.19 ms	2 - 222 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	1.881 ms	1 - 183 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® X2 Elite	2.315 ms	1 - 1 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® X Elite	4.894 ms	1 - 1 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	3.338 ms	0 - 220 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS6490	19.621 ms	1 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	13.258 ms	1 - 181 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	4.599 ms	1 - 2 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8775P	4.702 ms	1 - 184 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS9075	4.797 ms	1 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCM6690	81.041 ms	1 - 197 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	7.485 ms	1 - 218 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA7255P	13.258 ms	1 - 181 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8295P	7.746 ms	1 - 182 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	2.549 ms	0 - 180 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	7.866 ms	1 - 188 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	262.276 ms	0 - 244 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	304.396 ms	0 - 322 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8275 (Proxy)	721.465 ms	0 - 234 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8550 (Proxy)	404.823 ms	0 - 2 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8775P	386.033 ms	0 - 233 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS9075	396.158 ms	0 - 66 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8450 (Proxy)	443.425 ms	0 - 315 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA7255P	721.465 ms	0 - 234 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8295P	468.526 ms	0 - 228 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	260.795 ms	0 - 239 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	601.939 ms	0 - 444 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	577.435 ms	0 - 537 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS6490	1778.784 ms	267 - 428 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	1512.318 ms	0 - 443 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	812.034 ms	0 - 3 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8775P	796.513 ms	0 - 442 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS9075	571.08 ms	0 - 65 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCM6690	1601.488 ms	285 - 454 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	867.357 ms	1 - 490 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA7255P	1512.318 ms	0 - 443 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8295P	869.171 ms	0 - 440 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	494.44 ms	0 - 524 MB	NPU

License

• The license for the original implementation of ResNet-2Plus1D 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.