Swin-Tiny: Optimized for Qualcomm Devices

SwinTiny 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-Tiny 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-Tiny 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-Tiny 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: 28.8M
• Model size (float): 110 MB
• Model size (w8a16): 29.9 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
Swin-Tiny	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	4.071 ms	1 - 254 MB	NPU
Swin-Tiny	ONNX	float	Snapdragon® X2 Elite	4.398 ms	61 - 61 MB	NPU
Swin-Tiny	ONNX	float	Snapdragon® X Elite	10.97 ms	60 - 60 MB	NPU
Swin-Tiny	ONNX	float	Snapdragon® 8 Gen 3 Mobile	6.896 ms	0 - 315 MB	NPU
Swin-Tiny	ONNX	float	Qualcomm® QCS8550 (Proxy)	10.279 ms	0 - 74 MB	NPU
Swin-Tiny	ONNX	float	Qualcomm® QCS9075	12.34 ms	0 - 4 MB	NPU
Swin-Tiny	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	5.027 ms	1 - 216 MB	NPU
Swin-Tiny	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	3.432 ms	0 - 248 MB	NPU
Swin-Tiny	ONNX	w8a16	Snapdragon® X2 Elite	3.678 ms	33 - 33 MB	NPU
Swin-Tiny	ONNX	w8a16	Snapdragon® X Elite	8.831 ms	34 - 34 MB	NPU
Swin-Tiny	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	5.445 ms	0 - 348 MB	NPU
Swin-Tiny	ONNX	w8a16	Qualcomm® QCS6490	423.954 ms	96 - 112 MB	CPU
Swin-Tiny	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	8.315 ms	0 - 44 MB	NPU
Swin-Tiny	ONNX	w8a16	Qualcomm® QCS9075	10.077 ms	0 - 3 MB	NPU
Swin-Tiny	ONNX	w8a16	Qualcomm® QCM6690	217.014 ms	107 - 123 MB	CPU
Swin-Tiny	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	4.357 ms	0 - 289 MB	NPU
Swin-Tiny	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	199.597 ms	107 - 124 MB	CPU
Swin-Tiny	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	3.857 ms	0 - 424 MB	NPU
Swin-Tiny	QNN_DLC	float	Snapdragon® X2 Elite	4.413 ms	1 - 1 MB	NPU
Swin-Tiny	QNN_DLC	float	Snapdragon® X Elite	10.394 ms	1 - 1 MB	NPU
Swin-Tiny	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	6.309 ms	1 - 595 MB	NPU
Swin-Tiny	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	22.017 ms	1 - 440 MB	NPU
Swin-Tiny	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	9.609 ms	1 - 3 MB	NPU
Swin-Tiny	QNN_DLC	float	Qualcomm® SA8775P	10.939 ms	1 - 182 MB	NPU
Swin-Tiny	QNN_DLC	float	Qualcomm® QCS9075	11.859 ms	1 - 3 MB	NPU
Swin-Tiny	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	15.323 ms	0 - 244 MB	NPU
Swin-Tiny	QNN_DLC	float	Qualcomm® SA7255P	22.017 ms	1 - 440 MB	NPU
Swin-Tiny	QNN_DLC	float	Qualcomm® SA8295P	14.309 ms	1 - 435 MB	NPU
Swin-Tiny	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	4.774 ms	1 - 420 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	3.834 ms	0 - 442 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Snapdragon® X2 Elite	4.366 ms	0 - 0 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Snapdragon® X Elite	10.904 ms	0 - 0 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	6.546 ms	0 - 510 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	17.467 ms	0 - 437 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	10.008 ms	0 - 3 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Qualcomm® SA8775P	10.563 ms	0 - 446 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Qualcomm® QCS9075	11.958 ms	0 - 2 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Qualcomm® QCM6690	43.373 ms	0 - 518 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Qualcomm® SA7255P	17.467 ms	0 - 437 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	4.923 ms	0 - 434 MB	NPU
Swin-Tiny	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	10.622 ms	0 - 427 MB	NPU
Swin-Tiny	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	4.095 ms	0 - 203 MB	NPU
Swin-Tiny	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	6.861 ms	0 - 281 MB	NPU
Swin-Tiny	TFLITE	float	Qualcomm® QCS8275 (Proxy)	23.322 ms	0 - 209 MB	NPU
Swin-Tiny	TFLITE	float	Qualcomm® QCS8550 (Proxy)	10.613 ms	0 - 3 MB	NPU
Swin-Tiny	TFLITE	float	Qualcomm® SA8775P	11.662 ms	0 - 209 MB	NPU
Swin-Tiny	TFLITE	float	Qualcomm® QCS9075	12.887 ms	0 - 60 MB	NPU
Swin-Tiny	TFLITE	float	Qualcomm® QCS8450 (Proxy)	15.623 ms	0 - 266 MB	NPU
Swin-Tiny	TFLITE	float	Qualcomm® SA7255P	23.322 ms	0 - 209 MB	NPU
Swin-Tiny	TFLITE	float	Qualcomm® SA8295P	15.204 ms	0 - 208 MB	NPU
Swin-Tiny	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	5.146 ms	0 - 199 MB	NPU

License

• The license for the original implementation of Swin-Tiny 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.