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 2.45, ONNX Runtime 1.27.1	Download
QNN_DLC	float	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-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:

• Input resolution: 112x112
• Model checkpoint: Kinetics-400
• Model size (float): 120 MB
• Model size (w8a8): 30.8 MB
• Number of parameters: 31.5M

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
ResNet-2Plus1D	ONNX	float	Snapdragon® X2 Elite	9.208 ms	2 - 2 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® X Elite	12.157 ms	61 - 61 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Gen 3 Mobile	9.065 ms	0 - 277 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Gen 1 Mobile	25.777 ms	1 - 260 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	24.083 ms	2 - 8 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	12.019 ms	0 - 64 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® QCS8450	25.777 ms	1 - 260 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	21.415 ms	2 - 7 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	12.157 ms	61 - 61 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ Q-8750	7.173 ms	0 - 206 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Elite Mobile	7.173 ms	0 - 206 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	5.601 ms	1 - 209 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X2 Elite	6.631 ms	2 - 2 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X Elite	12.99 ms	2 - 2 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	9.183 ms	0 - 289 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	28.809 ms	0 - 269 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	23.838 ms	2 - 7 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	12.4 ms	2 - 4 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8775P	21.218 ms	1 - 216 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8650P	21.218 ms	1 - 216 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8255P	21.218 ms	1 - 216 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8450	28.809 ms	0 - 269 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	21.748 ms	2 - 6 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	12.99 ms	2 - 2 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	7.191 ms	2 - 220 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA7255P	81.665 ms	2 - 216 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8295P	22.751 ms	1 - 198 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite Mobile	7.191 ms	2 - 220 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	5.741 ms	2 - 229 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	287.301 ms	0 - 315 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	435.561 ms	0 - 301 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	364.198 ms	0 - 67 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	380.342 ms	0 - 3 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8775P	380.161 ms	0 - 237 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8650P	380.161 ms	0 - 237 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8255P	380.161 ms	0 - 237 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8450	435.561 ms	0 - 301 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	392.034 ms	0 - 66 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	265.447 ms	0 - 242 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA7255P	700.134 ms	0 - 238 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8295P	451.36 ms	0 - 230 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite Mobile	265.447 ms	0 - 242 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	256.81 ms	0 - 252 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	585.953 ms	0 - 513 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Gen 1 Mobile	861.392 ms	1 - 470 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ QCS6490	2969.326 ms	279 - 440 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ IQ-8275	758.929 ms	0 - 67 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	775.549 ms	0 - 3 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8775P	799.85 ms	1 - 440 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8650P	799.85 ms	1 - 440 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8255P	799.85 ms	1 - 440 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS8450	861.392 ms	1 - 470 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ IQ-9075	796.024 ms	0 - 66 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-6690	1605.584 ms	309 - 470 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-7790	1231.921 ms	296 - 369 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-8750	520.05 ms	0 - 456 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA7255P	1521.328 ms	1 - 439 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8295P	865.511 ms	1 - 437 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Elite Mobile	520.05 ms	0 - 456 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	639.093 ms	0 - 481 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	1231.921 ms	296 - 369 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.