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 2.45, ONNX Runtime 1.25.0	Download
ONNX	w8a16	Universal	QAIRT 2.45, ONNX Runtime 1.25.0	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® X2 Elite	6.809 ms	2 - 2 MB	NPU
Swin-Small	ONNX	float	Snapdragon® X Elite	16.21 ms	100 - 100 MB	NPU
Swin-Small	ONNX	float	Snapdragon® 8 Gen 3 Mobile	10.208 ms	1 - 426 MB	NPU
Swin-Small	ONNX	float	Snapdragon® 8 Gen 1 Mobile	23.988 ms	0 - 410 MB	NPU
Swin-Small	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	15.588 ms	0 - 515 MB	NPU
Swin-Small	ONNX	float	Qualcomm® QCS8450	23.988 ms	0 - 410 MB	NPU
Swin-Small	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	17.885 ms	0 - 4 MB	NPU
Swin-Small	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	6.421 ms	0 - 619 MB	NPU
Swin-Small	ONNX	float	Snapdragon® 8 Elite Mobile	7.837 ms	1 - 606 MB	NPU
Swin-Small	ONNX	float	Qualcomm® Dragonwing™ Q-8750	7.837 ms	1 - 606 MB	NPU
Swin-Small	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	16.21 ms	100 - 100 MB	NPU
Swin-Small	ONNX	w8a16	Snapdragon® X2 Elite	5.981 ms	1 - 1 MB	NPU
Swin-Small	ONNX	w8a16	Snapdragon® X Elite	13.83 ms	54 - 54 MB	NPU
Swin-Small	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	8.711 ms	0 - 529 MB	NPU
Swin-Small	ONNX	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	13.129 ms	0 - 61 MB	NPU
Swin-Small	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-9075	14.21 ms	0 - 3 MB	NPU
Swin-Small	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	14.115 ms	0 - 390 MB	NPU
Swin-Small	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	5.649 ms	0 - 453 MB	NPU
Swin-Small	ONNX	w8a16	Qualcomm® Dragonwing™ Q-6690	56.972 ms	0 - 575 MB	NPU
Swin-Small	ONNX	w8a16	Snapdragon® 8 Elite Mobile	6.976 ms	0 - 441 MB	NPU
Swin-Small	ONNX	w8a16	Qualcomm® Dragonwing™ Q-7790	14.115 ms	0 - 390 MB	NPU
Swin-Small	ONNX	w8a16	Qualcomm® Dragonwing™ Q-8750	6.976 ms	0 - 441 MB	NPU
Swin-Small	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-X7181	13.83 ms	54 - 54 MB	NPU
Swin-Small	QNN_DLC	float	Snapdragon® X2 Elite	7.084 ms	1 - 1 MB	NPU
Swin-Small	QNN_DLC	float	Snapdragon® X Elite	16.45 ms	1 - 1 MB	NPU
Swin-Small	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	9.965 ms	1 - 408 MB	NPU
Swin-Small	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	23.979 ms	0 - 397 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® QCS8275	37.922 ms	1 - 275 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	15.359 ms	1 - 3 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® SA8775P	17.398 ms	1 - 593 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® SA8650P	17.398 ms	1 - 593 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® SA8255P	17.398 ms	1 - 593 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® QCS8450	23.979 ms	0 - 397 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	22.492 ms	1 - 3 MB	NPU
Swin-Small	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	6.239 ms	0 - 602 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® SA7255P	37.922 ms	1 - 275 MB	NPU
Swin-Small	QNN_DLC	float	Snapdragon® 8 Elite Mobile	7.601 ms	1 - 587 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® SA8295P	22.631 ms	1 - 269 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	7.601 ms	1 - 587 MB	NPU
Swin-Small	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	16.45 ms	1 - 1 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® X2 Elite	7.239 ms	0 - 0 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® X Elite	17.275 ms	0 - 0 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	10.759 ms	0 - 872 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® QCS8275	29.266 ms	0 - 636 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	16.231 ms	0 - 272 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® SA8775P	16.726 ms	0 - 638 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® SA8650P	16.726 ms	0 - 638 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® SA8255P	16.726 ms	0 - 638 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-9075	23.292 ms	0 - 2 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	17.046 ms	0 - 652 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	6.464 ms	0 - 640 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-6690	72.958 ms	0 - 751 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® SA7255P	29.266 ms	0 - 636 MB	NPU
Swin-Small	QNN_DLC	w8a16	Snapdragon® 8 Elite Mobile	8.073 ms	0 - 621 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-7790	17.046 ms	0 - 652 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-8750	8.073 ms	0 - 621 MB	NPU
Swin-Small	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-X7181	17.275 ms	0 - 0 MB	NPU
Swin-Small	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	10.154 ms	0 - 430 MB	NPU
Swin-Small	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	23.37 ms	0 - 411 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® QCS8275	37.406 ms	0 - 306 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	15.431 ms	0 - 3 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® SA8775P	17.475 ms	0 - 306 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® SA8650P	17.475 ms	0 - 306 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® SA8255P	17.475 ms	0 - 306 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® QCS8450	23.37 ms	0 - 411 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	17.648 ms	0 - 104 MB	NPU
Swin-Small	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	6.605 ms	6 - 328 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® SA7255P	37.406 ms	0 - 306 MB	NPU
Swin-Small	TFLITE	float	Snapdragon® 8 Elite Mobile	7.841 ms	0 - 304 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® SA8295P	22.95 ms	0 - 297 MB	NPU
Swin-Small	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	7.841 ms	0 - 304 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.