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 1.25.0	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-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® X2 Elite	33.884 ms	201 - 201 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® X Elite	66.954 ms	170 - 170 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Gen 3 Mobile	52.837 ms	12 - 762 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Gen 1 Mobile	142.655 ms	12 - 590 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® QCS8550 (Proxy)	66.916 ms	12 - 15 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® QCS8450	142.655 ms	12 - 590 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Elite Mobile	41.99 ms	1 - 547 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	31.83 ms	1 - 563 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® QCS9075	110.635 ms	11 - 68 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® QCS8750	41.99 ms	1 - 547 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® QCS7181	66.954 ms	170 - 170 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® X2 Elite	11.915 ms	210 - 210 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® X Elite	25.073 ms	179 - 179 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	19.795 ms	3 - 458 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Gen 1 Mobile	44.175 ms	3 - 453 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	26.116 ms	0 - 41 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS8450	44.175 ms	3 - 453 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	11.961 ms	0 - 381 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	41.321 ms	3 - 383 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS9075	23.235 ms	3 - 48 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Elite Mobile	15.452 ms	2 - 372 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS7790	41.321 ms	3 - 383 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS8750	15.452 ms	2 - 372 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS7181	25.073 ms	179 - 179 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X2 Elite	34.378 ms	12 - 12 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X Elite	66.699 ms	12 - 12 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	51.825 ms	12 - 717 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	144.3 ms	4 - 593 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8275	424.679 ms	1 - 493 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	67.62 ms	12 - 15 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8450	144.3 ms	4 - 593 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite Mobile	42.075 ms	0 - 487 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8295P	118.43 ms	0 - 398 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	29.487 ms	12 - 509 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA7255P	424.679 ms	1 - 493 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS9075	112.074 ms	12 - 25 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8750	42.075 ms	0 - 487 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS7181	66.699 ms	12 - 12 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® X2 Elite	17.43 ms	3 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® X Elite	27.071 ms	3 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	21.462 ms	3 - 468 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Gen 1 Mobile	46.364 ms	3 - 459 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS6490	87.877 ms	5 - 9 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS8275	75.133 ms	2 - 373 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	26.852 ms	3 - 6 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS8450	46.364 ms	3 - 459 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	13.092 ms	3 - 390 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	46.383 ms	3 - 387 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS9075	23.942 ms	3 - 7 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA7255P	75.133 ms	2 - 373 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Elite Mobile	16.838 ms	3 - 374 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCM6690	440.107 ms	3 - 416 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8295P	41.916 ms	0 - 367 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS7790	46.383 ms	3 - 387 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS8750	16.838 ms	3 - 374 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS7181	27.071 ms	3 - 3 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1485.811 ms	0 - 795 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	2368.426 ms	0 - 737 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8275	3630.664 ms	1 - 581 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8550 (Proxy)	1929.978 ms	1 - 3 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8775P	3776.585 ms	1392 - 1408 MB	CPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8650P	3776.585 ms	1392 - 1408 MB	CPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8255P	3776.585 ms	1392 - 1408 MB	CPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8450	2368.426 ms	0 - 737 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite Mobile	1347.963 ms	0 - 572 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8295P	2463.021 ms	1 - 540 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1272.998 ms	0 - 574 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA7255P	3630.664 ms	1 - 581 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS9075	2052.77 ms	0 - 81 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8750	1347.963 ms	0 - 572 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	3088.485 ms	3 - 1623 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Gen 1 Mobile	4835.403 ms	3 - 1340 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS6490	8632.096 ms	1404 - 2196 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS8275	8124.22 ms	3 - 1383 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	4138.462 ms	1 - 5 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8775P	5023.53 ms	1404 - 1419 MB	CPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8650P	5023.53 ms	1404 - 1419 MB	CPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8255P	5023.53 ms	1404 - 1419 MB	CPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS8450	4835.403 ms	3 - 1340 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	3458.276 ms	0 - 1452 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	6137.62 ms	1363 - 1743 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS9075	2951.101 ms	0 - 74 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA7255P	8124.22 ms	3 - 1383 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Elite Mobile	2824.849 ms	1 - 1677 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8295P	4661.786 ms	2 - 1091 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS7790	6137.62 ms	1363 - 1743 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS8750	2824.849 ms	1 - 1677 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.