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
ONNX	w8a8	Universal	QAIRT 2.45, ONNX Runtime 1.27.1	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	33.888 ms	12 - 12 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® X Elite	66.932 ms	57 - 57 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Gen 3 Mobile	52.628 ms	4 - 766 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Gen 1 Mobile	141.817 ms	1 - 586 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	69.204 ms	0 - 71 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® QCS8450	141.817 ms	1 - 586 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	111.28 ms	11 - 26 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	66.932 ms	57 - 57 MB	NPU
ResNet-2Plus1D	ONNX	float	Qualcomm® Dragonwing™ Q-8750	42.336 ms	0 - 549 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Elite Mobile	42.336 ms	0 - 549 MB	NPU
ResNet-2Plus1D	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	32.148 ms	3 - 570 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® X2 Elite	11.93 ms	3 - 3 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® X Elite	25.02 ms	32 - 32 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	19.526 ms	0 - 468 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Gen 1 Mobile	44.772 ms	0 - 460 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	26.111 ms	0 - 471 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® QCS8450	44.772 ms	0 - 460 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-9075	23.847 ms	0 - 6 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ IQ-X7181	25.02 ms	32 - 32 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Qualcomm® Dragonwing™ Q-8750	15.767 ms	0 - 373 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Elite Mobile	15.767 ms	0 - 373 MB	NPU
ResNet-2Plus1D	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	12.029 ms	0 - 386 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X2 Elite	34.377 ms	12 - 12 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® X Elite	66.745 ms	12 - 12 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	51.95 ms	11 - 717 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	144.962 ms	3 - 595 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8275	424.692 ms	1 - 492 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	68.213 ms	12 - 857 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8775P	110.909 ms	1 - 490 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8650P	110.909 ms	1 - 490 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8255P	110.909 ms	1 - 490 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® QCS8450	144.962 ms	3 - 595 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	112.001 ms	12 - 25 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	66.745 ms	12 - 12 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	42.196 ms	0 - 489 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA7255P	424.692 ms	1 - 492 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Qualcomm® SA8295P	118.429 ms	0 - 398 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite Mobile	42.196 ms	0 - 489 MB	NPU
ResNet-2Plus1D	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	29.863 ms	12 - 509 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® X2 Elite	13.564 ms	3 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® X Elite	26.962 ms	3 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	21.45 ms	3 - 465 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Gen 1 Mobile	47.823 ms	3 - 461 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS6490	131.579 ms	2 - 6 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS8275	75.115 ms	2 - 372 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	27.969 ms	0 - 19 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8775P	24.465 ms	3 - 371 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8650P	24.465 ms	3 - 371 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8255P	24.465 ms	3 - 371 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® QCS8450	47.823 ms	3 - 461 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-9075	24.708 ms	0 - 5 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-X7181	26.962 ms	3 - 3 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-6690	435.241 ms	3 - 425 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-7790	47.007 ms	3 - 389 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-8750	16.967 ms	3 - 374 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA7255P	75.115 ms	2 - 372 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Qualcomm® SA8295P	41.916 ms	0 - 367 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Elite Mobile	16.967 ms	3 - 374 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	13.242 ms	3 - 390 MB	NPU
ResNet-2Plus1D	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	47.007 ms	3 - 389 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1500.788 ms	0 - 798 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	2377.92 ms	1 - 736 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1942.958 ms	0 - 4 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8775P	1944.202 ms	1 - 579 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8650P	1944.202 ms	1 - 579 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8255P	1944.202 ms	1 - 579 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® QCS8450	2377.92 ms	1 - 736 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	2008.742 ms	0 - 82 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	1410.823 ms	1 - 571 MB	NPU
ResNet-2Plus1D	TFLITE	float	Qualcomm® SA8295P	2444.504 ms	1 - 540 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite Mobile	1410.823 ms	1 - 571 MB	NPU
ResNet-2Plus1D	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1268.766 ms	0 - 573 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	3090.411 ms	1 - 1621 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Gen 1 Mobile	4717.092 ms	0 - 1336 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ QCS6490	8663.477 ms	1447 - 2239 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	4137.099 ms	2 - 6 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8775P	4293.165 ms	2 - 1380 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8650P	4293.165 ms	2 - 1380 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8255P	4293.165 ms	2 - 1380 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® QCS8450	4717.092 ms	0 - 1336 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ IQ-9075	4295.693 ms	0 - 79 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-6690	8068.589 ms	1393 - 1638 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-7790	6288.272 ms	1342 - 1725 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® Dragonwing™ Q-8750	2804.666 ms	1 - 1676 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Qualcomm® SA8295P	4657.152 ms	2 - 1091 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Elite Mobile	2804.666 ms	1 - 1676 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	3452.104 ms	2 - 1455 MB	NPU
ResNet-2Plus1D	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	6288.272 ms	1342 - 1725 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.