Swin-Small: Optimized for Qualcomm Devices

SwinSmall is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.

This is based on the implementation of Swin-Small 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 2.45	Download
QNN_DLC	w8a16	Universal	QAIRT 2.45	Download
TFLITE	float	Universal	QAIRT 2.45	Download

For more device-specific assets and performance metrics, visit Swin-Small 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 Swin-Small on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.image_classification

Model Stats:

• Model checkpoint: Imagenet
• Input resolution: 224x224
• Number of parameters: 50.4M
• Model size (float): 193 MB
• Model size (w8a16): 52.5 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
Swin-Small	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	7.215 ms	1 - 406 MB	NPU
Swin-Small	ONNX	float	Snapdragon® X2 Elite	8.033 ms	109 - 109 MB	NPU
Swin-Small	ONNX	float	Snapdragon® X Elite	19.461 ms	108 - 108 MB	NPU
Swin-Small	ONNX	float	Snapdragon® 8 Gen 3 Mobile	12.427 ms	0 - 527 MB	NPU
Swin-Small	ONNX	float	Qualcomm® QCS8550 (Proxy)	18.509 ms	0 - 123 MB	NPU
Swin-Small	ONNX	float	Qualcomm® QCS9075	22.286 ms	0 - 4 MB	NPU
Swin-Small	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	8.952 ms	1 - 361 MB	NPU
Swin-Small	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	5.998 ms	0 - 400 MB	NPU
Swin-Small	ONNX	w8a16	Snapdragon® X2 Elite	6.446 ms	59 - 59 MB	NPU
Swin-Small	ONNX	w8a16	Snapdragon® X Elite	14.759 ms	64 - 64 MB	NPU
Swin-Small	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	9.412 ms	0 - 573 MB	NPU
Swin-Small	ONNX	w8a16	Qualcomm® QCS6490	801.98 ms	99 - 113 MB	CPU
Swin-Small	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	14.084 ms	0 - 7 MB	NPU
Swin-Small	ONNX	w8a16	Qualcomm® QCS9075	18.095 ms	0 - 3 MB	NPU
Swin-Small	ONNX	w8a16	Qualcomm® QCM6690	385.253 ms	81 - 99 MB	CPU
Swin-Small	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	7.502 ms	0 - 474 MB	NPU
Swin-Small	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	360.716 ms	107 - 128 MB	CPU
Swin-Small	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	6.47 ms	0 - 606 MB	NPU
Swin-Small	QNN_DLC	float	Snapdragon® X2 Elite	7.232 ms	1 - 1 MB	NPU
Swin-Small	QNN_DLC	float	Snapdragon® X Elite	16.783 ms	1 - 1 MB	NPU
Swin-Small	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	10.518 ms	0 - 976 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	39.187 ms	1 - 631 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	16.15 ms	1 - 387 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® SA8775P	18.046 ms	1 - 626 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® QCS9075	20.327 ms	1 - 3 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	25.298 ms	0 - 409 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® SA7255P	39.187 ms	1 - 631 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® SA8295P	23.36 ms	1 - 599 MB	NPU
Swin-Small	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	7.956 ms	1 - 621 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	6.495 ms	0 - 655 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® X2 Elite	7.364 ms	0 - 0 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® X Elite	17.596 ms	0 - 0 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	10.856 ms	0 - 883 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	29.028 ms	0 - 651 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	16.51 ms	0 - 3 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® SA8775P	16.767 ms	0 - 656 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® QCS9075	19.739 ms	0 - 2 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® QCM6690	78.03 ms	0 - 619 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® SA7255P	29.028 ms	0 - 651 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	8.302 ms	0 - 642 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	17.296 ms	0 - 685 MB	NPU
Swin-Small	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	7.102 ms	0 - 320 MB	NPU
Swin-Small	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	11.638 ms	0 - 449 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® QCS8275 (Proxy)	41.087 ms	0 - 317 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® QCS8550 (Proxy)	17.623 ms	0 - 4 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® SA8775P	19.653 ms	0 - 316 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® QCS9075	22.581 ms	0 - 104 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® QCS8450 (Proxy)	25.957 ms	0 - 428 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® SA7255P	41.087 ms	0 - 317 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® SA8295P	25.431 ms	0 - 310 MB	NPU
Swin-Small	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	8.893 ms	0 - 309 MB	NPU

License

• The license for the original implementation of Swin-Small can be found here.

References

• Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
• 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.