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 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.19.1	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	456.524 ms	9 - 1085 MB	NPU
Video-MAE	ONNX	float	Snapdragon® X2 Elite	449.272 ms	187 - 187 MB	NPU
Video-MAE	ONNX	float	Snapdragon® X Elite	586.794 ms	187 - 187 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Gen 3 Mobile	368.651 ms	1 - 1260 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS8550 (Proxy)	566.739 ms	0 - 219 MB	NPU
Video-MAE	ONNX	float	Qualcomm® QCS9075	702.93 ms	9 - 21 MB	NPU
Video-MAE	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	390.742 ms	1 - 1006 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	320.662 ms	5 - 1071 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® X2 Elite	320.484 ms	100 - 100 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® X Elite	477.996 ms	98 - 98 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	461.219 ms	5 - 1266 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® QCS6490	8924.347 ms	337 - 353 MB	CPU
Video-MAE	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	465.061 ms	0 - 128 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® QCS9075	1393.064 ms	0 - 7 MB	NPU
Video-MAE	ONNX	w8a16	Qualcomm® QCM6690	4490.407 ms	344 - 355 MB	CPU
Video-MAE	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	276.519 ms	1 - 1046 MB	NPU
Video-MAE	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	4388.447 ms	313 - 326 MB	CPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	330.03 ms	9 - 963 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X2 Elite	296.63 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® X Elite	472.162 ms	9 - 9 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	382.065 ms	9 - 1167 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	1090.226 ms	0 - 947 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	450.843 ms	9 - 11 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8775P	496.616 ms	0 - 972 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS9075	514.726 ms	11 - 22 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	580.84 ms	8 - 1064 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA7255P	1090.226 ms	0 - 947 MB	NPU
Video-MAE	QNN_DLC	float	Qualcomm® SA8295P	568.854 ms	0 - 858 MB	NPU
Video-MAE	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	260.148 ms	9 - 969 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	51.847 ms	0 - 934 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	92.345 ms	0 - 1132 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8275 (Proxy)	421.709 ms	0 - 923 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8550 (Proxy)	127.723 ms	0 - 237 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8775P	152.797 ms	0 - 924 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS9075	157.616 ms	0 - 206 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® QCS8450 (Proxy)	262.805 ms	0 - 1068 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA7255P	421.709 ms	0 - 923 MB	NPU
Video-MAE	TFLITE	float	Qualcomm® SA8295P	202.006 ms	0 - 867 MB	NPU
Video-MAE	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	69.86 ms	0 - 930 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.