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 1.24.1	Download
ONNX	w8a16	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
QNN_DLC	w8a16	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.17.0	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® X2 Elite	465.083 ms	188 - 188 MB	NPU
Video-MAE	ONNX	float	Snapdragon® X Elite	668.776 ms	188 - 188 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Gen 3 Mobile	388.833 ms	9 - 1257 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS8550 (Proxy)	644.907 ms	0 - 219 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS9075	686.387 ms	9 - 21 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	394.489 ms	0 - 979 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	444.692 ms	9 - 1065 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® X2 Elite	323.006 ms	123 - 123 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® X Elite	489.095 ms	129 - 129 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	447.917 ms	5 - 1858 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® QCS6490	9057.031 ms	363 - 381 MB	CPU
Video-MAE	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	473.026 ms	0 - 160 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® QCM6690	4452.468 ms	326 - 338 MB	CPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	278.713 ms	0 - 1538 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	4377.307 ms	376 - 388 MB	CPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	323.42 ms	5 - 1549 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X2 Elite	249.277 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X Elite	1308.348 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	414.95 ms	9 - 1110 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	2499.681 ms	1 - 889 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	1249.753 ms	9 - 12 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8775P	6237.053 ms	1 - 889 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS9075	519.754 ms	9 - 20 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	585.147 ms	9 - 1025 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA7255P	2499.681 ms	1 - 889 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8295P	655.834 ms	0 - 843 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	307.435 ms	9 - 934 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	322.365 ms	9 - 943 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	103.807 ms	0 - 1114 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8275 (Proxy)	428.414 ms	0 - 918 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8550 (Proxy)	141.755 ms	0 - 3 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8775P	164.5 ms	0 - 921 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS9075	176.822 ms	0 - 207 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8450 (Proxy)	302.041 ms	0 - 1056 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA7255P	428.414 ms	0 - 918 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8295P	216.681 ms	0 - 872 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	81.01 ms	0 - 923 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	63.809 ms	0 - 941 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.