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 1.25.0	Download
QNN_DLC	float	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® X2 Elite	1573.648 ms	167 - 167 MB	NPU
Video-MAE	ONNX	float	Snapdragon® X Elite	2876.574 ms	193 - 193 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS8550 (Proxy)	2818.888 ms	1 - 213 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite Mobile	1649.301 ms	4 - 5496 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1516.515 ms	1 - 5696 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS9075	5020.553 ms	46 - 137 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS8750	1649.301 ms	4 - 5496 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS7181	2876.574 ms	193 - 193 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X2 Elite	1968.594 ms	46 - 46 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X Elite	3275.201 ms	46 - 46 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	5135.131 ms	46 - 50 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite Mobile	3552.984 ms	1 - 6093 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8295P	3812.661 ms	36 - 5739 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	4070.911 ms	3 - 6272 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS9075	5458.899 ms	46 - 94 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8750	3552.984 ms	1 - 6093 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS7181	3275.201 ms	46 - 46 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8550 (Proxy)	5157.258 ms	1 - 5 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8775P	14234.218 ms	188 - 197 MB	CPU
Video-MAE	TFLITE	float	Qualcomm® SA8650P	14234.218 ms	188 - 197 MB	CPU
Video-MAE	TFLITE	float	Qualcomm® SA8255P	14234.218 ms	188 - 197 MB	CPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite Mobile	3536.957 ms	1 - 5953 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8295P	3741.683 ms	1 - 5603 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	4086.474 ms	1 - 6105 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS9075	5389.455 ms	0 - 280 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8750	3536.957 ms	1 - 5953 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.