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 2.45, ONNX Runtime 1.27.1	Download
ONNX	w8a16	Universal	QAIRT 2.45, ONNX Runtime 1.27.1	Download
QNN_DLC	float	Universal	QAIRT 2.45	Download
QNN_DLC	w8a16	Universal	QAIRT 2.45	Download
TFLITE	float	Universal	QAIRT 2.45	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® X2 Elite	465.657 ms	9 - 9 MB	NPU
Video-MAE	ONNX	float	Snapdragon® X Elite	615.161 ms	188 - 188 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Gen 3 Mobile	399.949 ms	1 - 1085 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Gen 1 Mobile	1390.512 ms	11 - 955 MB	NPU
Video-MAE	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	558.465 ms	9 - 22 MB	NPU
Video-MAE	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	594.644 ms	0 - 217 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS8450	1390.512 ms	11 - 955 MB	NPU
Video-MAE	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	560.516 ms	9 - 21 MB	NPU
Video-MAE	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	615.161 ms	188 - 188 MB	NPU
Video-MAE	ONNX	float	Qualcomm® Dragonwing™ Q-8750	379.001 ms	1 - 787 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite Mobile	379.001 ms	1 - 787 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	469.425 ms	4 - 795 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® X2 Elite	328.077 ms	5 - 5 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® X Elite	494.636 ms	100 - 100 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	371.206 ms	0 - 1091 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Gen 1 Mobile	1448.884 ms	5 - 1156 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ QCS6490	2319.296 ms	0 - 7 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-8275	544.73 ms	1 - 9 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	487.645 ms	0 - 111 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® QCS8450	1448.884 ms	5 - 1156 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-9075	487.856 ms	0 - 7 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-X7181	494.636 ms	100 - 100 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ Q-7790	1079.643 ms	5 - 1356 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® Dragonwing™ Q-8750	290.863 ms	2 - 875 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Elite Mobile	290.863 ms	2 - 875 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	362.553 ms	1 - 897 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	1079.643 ms	5 - 1356 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X2 Elite	255.624 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X Elite	470.017 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	383.664 ms	10 - 986 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	562.401 ms	9 - 898 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	519.983 ms	9 - 21 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	456.357 ms	10 - 12 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8775P	480.128 ms	0 - 679 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8650P	480.128 ms	0 - 679 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8255P	480.128 ms	0 - 679 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8450	562.401 ms	9 - 898 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	476.202 ms	9 - 20 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	470.017 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	266.012 ms	7 - 684 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA7255P	1125.532 ms	3 - 680 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8295P	616.266 ms	1 - 651 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite Mobile	266.012 ms	7 - 684 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	291.16 ms	9 - 702 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® X2 Elite	336.034 ms	5 - 5 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® X Elite	518.054 ms	5 - 5 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	378.224 ms	5 - 1247 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-8275	519.134 ms	5 - 12 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	511.946 ms	5 - 7 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® SA8775P	514.977 ms	2 - 798 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® SA8650P	514.977 ms	2 - 798 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® SA8255P	514.977 ms	2 - 798 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-9075	508.939 ms	3 - 9 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-X7181	518.054 ms	5 - 5 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-6690	3549.028 ms	5 - 1342 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-7790	713.062 ms	5 - 1154 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-8750	292.97 ms	5 - 782 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® SA7255P	980.449 ms	1 - 832 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 8 Elite Mobile	292.97 ms	5 - 782 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	334.391 ms	5 - 809 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	713.062 ms	5 - 1154 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	124.051 ms	0 - 977 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	260.283 ms	1 - 935 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	184.639 ms	0 - 208 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	158.207 ms	0 - 4 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8775P	181.366 ms	0 - 709 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8650P	181.366 ms	0 - 709 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8255P	181.366 ms	0 - 709 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8450	260.283 ms	1 - 935 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	178.838 ms	0 - 207 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	91.622 ms	0 - 688 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA7255P	469.667 ms	1 - 681 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8295P	239.007 ms	0 - 649 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite Mobile	91.622 ms	0 - 688 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	63.749 ms	0 - 729 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.