CavaFace: Optimized for Qualcomm Devices

A PyTorch-based framework for training face recognition models that generates facial embeddings for verification and identification tasks

This is based on the implementation of CavaFace 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
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 CavaFace 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 CavaFace on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.object_detection

Model Stats:

• Model checkpoint: IR_SE_100_Combined_Epoch_24.pt
• Input resolution: 112x112
• Number of parameters: 65.5M
• Model size (float): 249.96MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
CavaFace	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	2.267 ms	0 - 92 MB	NPU
CavaFace	ONNX	float	Snapdragon® X2 Elite	2.354 ms	126 - 126 MB	NPU
CavaFace	ONNX	float	Snapdragon® X Elite	4.499 ms	126 - 126 MB	NPU
CavaFace	ONNX	float	Snapdragon® 8 Gen 3 Mobile	3.194 ms	0 - 111 MB	NPU
CavaFace	ONNX	float	Qualcomm® QCS8550 (Proxy)	4.343 ms	0 - 131 MB	NPU
CavaFace	ONNX	float	Qualcomm® QCS9075	6.788 ms	0 - 3 MB	NPU
CavaFace	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	2.67 ms	0 - 91 MB	NPU
CavaFace	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	2.246 ms	0 - 84 MB	NPU
CavaFace	QNN_DLC	float	Snapdragon® X2 Elite	2.591 ms	0 - 0 MB	NPU
CavaFace	QNN_DLC	float	Snapdragon® X Elite	4.462 ms	0 - 0 MB	NPU
CavaFace	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	3.191 ms	0 - 103 MB	NPU
CavaFace	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	24.698 ms	0 - 81 MB	NPU
CavaFace	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	4.308 ms	0 - 244 MB	NPU
CavaFace	QNN_DLC	float	Qualcomm® SA8775P	6.933 ms	0 - 81 MB	NPU
CavaFace	QNN_DLC	float	Qualcomm® QCS9075	6.771 ms	0 - 2 MB	NPU
CavaFace	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	8.856 ms	0 - 110 MB	NPU
CavaFace	QNN_DLC	float	Qualcomm® SA7255P	24.698 ms	0 - 81 MB	NPU
CavaFace	QNN_DLC	float	Qualcomm® SA8295P	7.964 ms	0 - 85 MB	NPU
CavaFace	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	2.615 ms	0 - 82 MB	NPU
CavaFace	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	2.223 ms	0 - 101 MB	NPU
CavaFace	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	3.152 ms	0 - 214 MB	NPU
CavaFace	TFLITE	float	Qualcomm® QCS8275 (Proxy)	24.566 ms	0 - 98 MB	NPU
CavaFace	TFLITE	float	Qualcomm® QCS8550 (Proxy)	4.241 ms	0 - 3 MB	NPU
CavaFace	TFLITE	float	Qualcomm® SA8775P	6.916 ms	0 - 98 MB	NPU
CavaFace	TFLITE	float	Qualcomm® QCS9075	6.704 ms	0 - 128 MB	NPU
CavaFace	TFLITE	float	Qualcomm® QCS8450 (Proxy)	8.796 ms	0 - 218 MB	NPU
CavaFace	TFLITE	float	Qualcomm® SA7255P	24.566 ms	0 - 98 MB	NPU
CavaFace	TFLITE	float	Qualcomm® SA8295P	7.939 ms	0 - 100 MB	NPU
CavaFace	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	2.603 ms	0 - 102 MB	NPU

License

• The license for the original implementation of CavaFace can be found here.

References

• 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.