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 2.42, ONNX Runtime 1.24.3	Download
ONNX	w8a16	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	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:

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

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 Mobile	448.715 ms	9 - 997 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite Mobile	380.738 ms	3 - 915 MB	NPU
Video-MAE	ONNX	float	Snapdragon® X2 Elite	438.207 ms	186 - 186 MB	NPU
Video-MAE	ONNX	float	Snapdragon® X Elite	665.106 ms	187 - 187 MB	NPU
Video-MAE	ONNX	float	Snapdragon® X Elite	665.106 ms	187 - 187 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Gen 3 Mobile	401.658 ms	20 - 1209 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS8550 (Proxy)	632.18 ms	0 - 220 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS9075	686.02 ms	9 - 21 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	380.738 ms	3 - 915 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	318.12 ms	5 - 1026 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Elite Mobile	268.796 ms	1 - 970 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® X2 Elite	315.424 ms	99 - 99 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® X Elite	460.275 ms	99 - 99 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® X Elite	460.275 ms	99 - 99 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	420.106 ms	5 - 1368 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® QCS6490	9558.694 ms	475 - 491 MB	CPU
Video-MAE	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	444.439 ms	0 - 112 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® QCS9075	573.657 ms	0 - 7 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® QCM6690	4761.175 ms	439 - 450 MB	CPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	268.796 ms	1 - 970 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	4610.601 ms	481 - 494 MB	CPU
Video-MAE	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	4610.601 ms	481 - 494 MB	CPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	329.469 ms	9 - 722 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite Mobile	261.13 ms	9 - 750 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X2 Elite	325.489 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X Elite	459.949 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X Elite	459.949 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	344.282 ms	9 - 907 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	1119.507 ms	0 - 773 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	439.851 ms	9 - 11 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8775P	455.299 ms	1 - 661 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8775P	455.299 ms	1 - 661 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8775P	455.299 ms	1 - 661 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS9075	487.037 ms	9 - 20 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	545.485 ms	9 - 850 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA7255P	1119.507 ms	0 - 773 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8295P	558.95 ms	0 - 597 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	261.13 ms	9 - 750 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	295.73 ms	5 - 803 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 8 Elite Mobile	239.763 ms	5 - 792 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® X2 Elite	294.73 ms	5 - 5 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® X Elite	452.903 ms	5 - 5 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® X Elite	452.903 ms	5 - 5 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	299.861 ms	5 - 955 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	946.992 ms	1 - 831 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	432.828 ms	5 - 7 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® SA8775P	441.17 ms	1 - 807 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® SA8775P	441.17 ms	1 - 807 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® SA8775P	441.17 ms	1 - 807 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® QCS9075	415.886 ms	1 - 7 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® QCM6690	2413.695 ms	5 - 1311 MB	NPU
Video-MAE	QNN_DLC	w8a16	Qualcomm® SA7255P	946.992 ms	1 - 831 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	239.763 ms	5 - 792 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	707.904 ms	5 - 1070 MB	NPU
Video-MAE	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	707.904 ms	5 - 1070 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	53.713 ms	0 - 717 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite Mobile	64.725 ms	0 - 711 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	92.991 ms	0 - 965 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8275 (Proxy)	416.588 ms	0 - 724 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8550 (Proxy)	125.384 ms	0 - 4 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8775P	148.248 ms	0 - 712 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8775P	148.248 ms	0 - 712 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8775P	148.248 ms	0 - 712 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS9075	158.52 ms	0 - 206 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8450 (Proxy)	265.013 ms	0 - 897 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA7255P	416.588 ms	0 - 724 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8295P	202.193 ms	0 - 656 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	64.725 ms	0 - 711 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.