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	34.069 ms	12 - 12 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® X Elite	67.09 ms	57 - 57 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Gen 3 Mobile	52.889 ms	4 - 767 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Gen 1 Mobile	141.115 ms	0 - 587 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	68.101 ms	0 - 512 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® QCS8450	141.115 ms	0 - 587 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	116.369 ms	11 - 26 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	31.749 ms	0 - 567 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Elite Mobile	42.553 ms	2 - 551 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ Q-8750	42.553 ms	2 - 551 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	67.09 ms	57 - 57 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® X2 Elite	11.835 ms	3 - 3 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® X Elite	25.065 ms	32 - 32 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	19.574 ms	0 - 470 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Gen 1 Mobile	45.037 ms	0 - 460 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	25.113 ms	0 - 40 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS8450	45.037 ms	0 - 460 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Elite Mobile	15.672 ms	0 - 372 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	41.03 ms	3 - 385 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	12.049 ms	0 - 386 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-9075	24.168 ms	0 - 6 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ Q-7790	41.03 ms	3 - 385 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ Q-8750	15.672 ms	0 - 372 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-X7181	25.065 ms	32 - 32 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X2 Elite	34.435 ms	12 - 12 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X Elite	66.649 ms	12 - 12 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	52.432 ms	12 - 720 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	144.24 ms	12 - 600 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8275	424.933 ms	1 - 492 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	66.538 ms	12 - 14 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8775P	110.962 ms	0 - 489 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8650P	110.962 ms	0 - 489 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8255P	110.962 ms	0 - 489 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8450	144.24 ms	12 - 600 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	111.628 ms	12 - 25 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	29.838 ms	5 - 501 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA7255P	424.933 ms	1 - 492 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite Mobile	41.822 ms	0 - 487 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8295P	118.422 ms	0 - 398 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	41.822 ms	0 - 487 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	66.649 ms	12 - 12 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® X2 Elite	13.127 ms	3 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® X Elite	27.033 ms	3 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	21.128 ms	3 - 468 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Gen 1 Mobile	47.629 ms	3 - 460 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS6490	131.287 ms	3 - 7 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS8275	75.159 ms	3 - 373 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	27.425 ms	3 - 421 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8775P	24.431 ms	3 - 371 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8650P	24.431 ms	3 - 371 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8255P	24.431 ms	3 - 371 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS8450	47.629 ms	3 - 460 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Elite Mobile	16.749 ms	3 - 373 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	46.667 ms	3 - 385 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	13.202 ms	3 - 390 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-9075	24.606 ms	3 - 7 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-6690	422.562 ms	3 - 424 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA7255P	75.159 ms	3 - 373 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8295P	41.911 ms	2 - 368 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-7790	46.667 ms	3 - 385 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-8750	16.749 ms	3 - 373 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-X7181	27.033 ms	3 - 3 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1492.431 ms	0 - 799 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	2376.848 ms	0 - 737 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8275	3622.265 ms	2 - 580 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1958.497 ms	0 - 4 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8775P	1961.124 ms	1 - 580 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8650P	1961.124 ms	1 - 580 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8255P	1961.124 ms	1 - 580 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8450	2376.848 ms	0 - 737 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	3772.093 ms	0 - 82 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1273.491 ms	0 - 573 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA7255P	3622.265 ms	2 - 580 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite Mobile	1414.464 ms	0 - 577 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8295P	2443.499 ms	1 - 541 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	1414.464 ms	0 - 577 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	3095.516 ms	1 - 1624 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Gen 1 Mobile	4758.684 ms	4 - 1343 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ QCS6490	8645.45 ms	1448 - 2240 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS8275	8118.17 ms	3 - 1383 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	4131.917 ms	1 - 4 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8775P	4291.613 ms	2 - 1381 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8650P	4291.613 ms	2 - 1381 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8255P	4291.613 ms	2 - 1381 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS8450	4758.684 ms	4 - 1343 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Elite Mobile	2860.879 ms	1 - 1681 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	6174.478 ms	1345 - 1727 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	3458.507 ms	0 - 1453 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ IQ-9075	4297.173 ms	0 - 79 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-6690	7966.761 ms	1382 - 1625 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA7255P	8118.17 ms	3 - 1383 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8295P	4657.251 ms	2 - 1090 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-7790	6174.478 ms	1345 - 1727 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-8750	2860.879 ms	1 - 1681 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.