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 2.50, ONNX Runtime 1.27.1	Download
ONNX	w8a8	Universal	QAIRT 2.50, ONNX Runtime 1.27.1	Download
QNN_DLC	float	Universal	QAIRT 2.50	Download
QNN_DLC	w8a8	Universal	QAIRT 2.50	Download
TFLITE	float	Universal	QAIRT 2.50	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® 8 Elite Gen 5 For Galaxy Mobile	5.804 ms	2 - 200 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	7.047 ms	0 - 193 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® X2 Elite	6.152 ms	2 - 2 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® X Elite	11.871 ms	61 - 61 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Gen 3 Mobile	9.161 ms	2 - 274 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Gen 1 Mobile	20.888 ms	2 - 252 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	23.82 ms	2 - 9 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	11.856 ms	0 - 64 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® QCS8450	20.888 ms	2 - 252 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	21.04 ms	2 - 8 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	11.871 ms	61 - 61 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ Q-8750	7.047 ms	0 - 193 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	1.957 ms	0 - 178 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	2.515 ms	0 - 174 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® X2 Elite	2.025 ms	2 - 2 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® X Elite	4.239 ms	31 - 31 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	3.258 ms	0 - 209 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Gen 1 Mobile	5.792 ms	1 - 212 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ QCS6490	18.038 ms	1 - 4 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-8275	4.383 ms	0 - 4 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	4.205 ms	0 - 34 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS8450	5.792 ms	1 - 212 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-9075	4.085 ms	1 - 4 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-X7181	4.239 ms	31 - 31 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ Q-6690	79.726 ms	0 - 188 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ Q-7790	7.706 ms	1 - 183 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ Q-8750	2.515 ms	0 - 174 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	7.706 ms	1 - 183 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	5.791 ms	2 - 200 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	7.149 ms	0 - 195 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X2 Elite	6.668 ms	2 - 2 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X Elite	12.268 ms	2 - 2 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	9.17 ms	0 - 269 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	20.688 ms	1 - 250 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	23.732 ms	2 - 7 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	11.971 ms	2 - 4 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8775P	20.74 ms	2 - 196 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8650P	20.74 ms	2 - 196 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8255P	20.74 ms	2 - 196 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8450	20.688 ms	1 - 250 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	21.197 ms	2 - 7 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	12.268 ms	2 - 2 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	7.149 ms	0 - 195 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8295P	22.226 ms	0 - 180 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	2.031 ms	1 - 177 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	2.62 ms	0 - 175 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® X2 Elite	2.436 ms	1 - 1 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® X Elite	4.579 ms	1 - 1 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	3.376 ms	1 - 208 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Gen 1 Mobile	5.752 ms	1 - 209 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS6490	18.266 ms	1 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-8275	4.439 ms	0 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	4.309 ms	1 - 2 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8775P	4.352 ms	1 - 173 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8650P	4.352 ms	1 - 173 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8255P	4.352 ms	1 - 173 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS8450	5.752 ms	1 - 209 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-9075	4.183 ms	1 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-X7181	4.579 ms	1 - 1 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-6690	78.904 ms	1 - 185 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-7790	7.897 ms	1 - 175 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-8750	2.62 ms	0 - 175 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8295P	7.318 ms	1 - 176 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	7.897 ms	1 - 175 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	229.977 ms	0 - 233 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	232.112 ms	0 - 224 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	240.561 ms	0 - 296 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	378.251 ms	1 - 291 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	330.057 ms	0 - 68 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	330.442 ms	0 - 2 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8450	378.251 ms	1 - 291 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	345.646 ms	0 - 67 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	232.112 ms	0 - 224 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8295P	416.851 ms	0 - 214 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.