Video-MAE: Optimized for Qualcomm Devices

Video MAE (Masked Auto Encoder) is a network for doing video classification that uses the ViT (Vision Transformer) backbone.

This is based on the implementation of Video-MAE 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	w8a16	Universal	QAIRT 2.50, ONNX Runtime 1.27.1	Download
QNN_DLC	float	Universal	QAIRT 2.50	Download
QNN_DLC	w8a16	Universal	QAIRT 2.50	Download
TFLITE	float	Universal	QAIRT 2.50	Download

For more device-specific assets and performance metrics, visit Video-MAE 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 Video-MAE on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.video_classification

Model Stats:

• Input resolution: 224x224
• Model checkpoint: Kinectics-400
• Model size (float): 335 MB
• Number of parameters: 87.7M

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
Video-MAE	ONNX	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	261.663 ms	3 - 753 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	299.359 ms	1 - 711 MB	NPU
Video-MAE	ONNX	float	Snapdragon® X2 Elite	292.882 ms	9 - 9 MB	NPU
Video-MAE	ONNX	float	Snapdragon® X Elite	527.533 ms	186 - 186 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Gen 3 Mobile	358.011 ms	3 - 1019 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Gen 1 Mobile	749.446 ms	10 - 893 MB	NPU
Video-MAE	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	504.382 ms	9 - 22 MB	NPU
Video-MAE	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	510.816 ms	0 - 217 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS8450	749.446 ms	10 - 893 MB	NPU
Video-MAE	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	545.976 ms	9 - 22 MB	NPU
Video-MAE	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	527.533 ms	186 - 186 MB	NPU
Video-MAE	ONNX	float	Qualcomm® Dragonwing™ Q-8750	299.359 ms	1 - 711 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	320.827 ms	2 - 813 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	284.676 ms	1 - 792 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® X2 Elite	312.615 ms	5 - 5 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® X Elite	500.628 ms	99 - 99 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	381.432 ms	0 - 1350 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Gen 1 Mobile	680.059 ms	7 - 1123 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ QCS6490	1682.595 ms	0 - 8 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-8275	470.041 ms	1 - 10 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	485.189 ms	0 - 110 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® QCS8450	680.059 ms	7 - 1123 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-9075	487.171 ms	0 - 8 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-X7181	500.628 ms	99 - 99 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ Q-6690	3827.066 ms	0 - 1381 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ Q-7790	625.614 ms	5 - 985 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ Q-8750	284.676 ms	1 - 792 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	625.614 ms	5 - 985 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	258.449 ms	9 - 687 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	269.852 ms	9 - 588 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X2 Elite	277.61 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X Elite	507.518 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	361.693 ms	9 - 865 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	712.271 ms	9 - 827 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	492.512 ms	9 - 21 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	501.37 ms	9 - 946 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8775P	521.34 ms	1 - 605 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8650P	521.34 ms	1 - 605 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8255P	521.34 ms	1 - 605 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8450	712.271 ms	9 - 827 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	516.224 ms	9 - 20 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	507.518 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	269.852 ms	9 - 588 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8295P	703.386 ms	1 - 544 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	532.121 ms	5 - 727 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	294.293 ms	5 - 709 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® X2 Elite	658.555 ms	5 - 5 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® X Elite	486.052 ms	5 - 5 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	502.324 ms	5 - 959 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-8275	602.801 ms	5 - 12 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	483.308 ms	5 - 7 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® SA8775P	618.647 ms	2 - 752 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® SA8650P	618.647 ms	2 - 752 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® SA8255P	618.647 ms	2 - 752 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-9075	612.88 ms	2 - 9 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-X7181	486.052 ms	5 - 5 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-7790	746.354 ms	5 - 887 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-8750	294.293 ms	5 - 709 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	746.354 ms	5 - 887 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	59.933 ms	0 - 656 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	83.944 ms	0 - 589 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	101.822 ms	0 - 1011 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	208.989 ms	1 - 839 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	162.07 ms	0 - 207 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	139.528 ms	0 - 3 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8775P	161.342 ms	0 - 573 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8650P	161.342 ms	0 - 573 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8255P	161.342 ms	0 - 573 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8450	208.989 ms	1 - 839 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	159.987 ms	0 - 206 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	83.944 ms	0 - 589 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA7255P	4032.841 ms	47 - 64 MB	CPU
Video-MAE	TFLITE	float	Qualcomm® SA8295P	209.836 ms	0 - 671 MB	NPU

License

• The license for the original implementation of Video-MAE can be found here.

References

• Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training
• 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.