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 2.37, ONNX Runtime 1.23.0	Download
ONNX	w8a16	Universal	QAIRT 2.37, ONNX Runtime 1.23.0	Download
QNN_DLC	float	Universal	QAIRT 2.42	Download
QNN_DLC	w8a16	Universal	QAIRT 2.42	Download
TFLITE	float	Universal	QAIRT 2.42, 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® X Elite	599.027 ms	187 - 187 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Gen 3 Mobile	417.051 ms	10 - 1193 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS8550 (Proxy)	572.814 ms	0 - 217 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS9075	672.39 ms	9 - 21 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	439.593 ms	1 - 971 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	594.011 ms	1 - 977 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X Elite	473.072 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	354.537 ms	9 - 1124 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	1038.798 ms	0 - 914 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	454.586 ms	9 - 12 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8775P	485.216 ms	0 - 956 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS9075	515.489 ms	11 - 22 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	579.173 ms	9 - 1066 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA7255P	1038.798 ms	0 - 914 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8295P	565.13 ms	0 - 858 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	331.251 ms	9 - 928 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	249.902 ms	11 - 966 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	100.537 ms	0 - 1169 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8275 (Proxy)	5536.906 ms	42 - 59 MB	CPU
Video-MAE	TFLITE	float	Qualcomm® QCS8550 (Proxy)	138.298 ms	0 - 4 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8775P	160.51 ms	0 - 961 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS9075	172.102 ms	0 - 207 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8450 (Proxy)	298.251 ms	0 - 1115 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA7255P	5536.906 ms	42 - 59 MB	CPU
Video-MAE	TFLITE	float	Qualcomm® SA8295P	214.16 ms	0 - 912 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	74.663 ms	0 - 962 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	60.822 ms	0 - 954 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.