SINet: Optimized for Qualcomm Devices

SINet is a machine learning model that is designed to segment people from close-up portrait images in real time.

This is based on the implementation of SINet 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 1.27.1	Download
QNN_DLC	float	Universal	QAIRT 2.45	Download
QNN_DLC	w8a16	Universal	QAIRT 2.45	Download
QNN_DLC	w8a8	Universal	QAIRT 2.45	Download
TFLITE	float	Universal	QAIRT 2.45	Download

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

Model Details

Model Type: Model_use_case.semantic_segmentation

Model Stats:

Model checkpoint: SINet.pth
Input resolution: 224x224
Number of output classes: 2 (foreground / background)
Number of parameters: 91.9K
Model size (float): 415 KB
Model size (w8a8): 241 KB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
SINet	ONNX	float	Snapdragon® X2 Elite	0.757 ms	2 - 2 MB	NPU
SINet	ONNX	float	Snapdragon® X Elite	1.76 ms	2 - 2 MB	NPU
SINet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.122 ms	0 - 47 MB	NPU
SINet	ONNX	float	Snapdragon® 8 Gen 1 Mobile	2.027 ms	1 - 57 MB	NPU
SINet	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.684 ms	0 - 3 MB	NPU
SINet	ONNX	float	Qualcomm® QCS8450	2.027 ms	1 - 57 MB	NPU
SINet	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	2.288 ms	1 - 5 MB	NPU
SINet	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	1.76 ms	2 - 2 MB	NPU
SINet	ONNX	float	Qualcomm® Dragonwing™ Q-8750	0.838 ms	0 - 33 MB	NPU
SINet	ONNX	float	Snapdragon® 8 Elite Mobile	0.838 ms	0 - 33 MB	NPU
SINet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.709 ms	0 - 38 MB	NPU
SINet	QNN_DLC	float	Snapdragon® X2 Elite	1.398 ms	1 - 1 MB	NPU
SINet	QNN_DLC	float	Snapdragon® X Elite	3.824 ms	1 - 1 MB	NPU
SINet	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	2.114 ms	0 - 46 MB	NPU
SINet	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	3.739 ms	0 - 55 MB	NPU
SINet	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8275	6.614 ms	1 - 34 MB	NPU
SINet	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	3.432 ms	1 - 2 MB	NPU
SINet	QNN_DLC	float	Qualcomm® SA8775P	3.885 ms	1 - 36 MB	NPU
SINet	QNN_DLC	float	Qualcomm® SA8650P	3.885 ms	1 - 36 MB	NPU
SINet	QNN_DLC	float	Qualcomm® SA8255P	3.885 ms	1 - 36 MB	NPU
SINet	QNN_DLC	float	Qualcomm® QCS8450	3.739 ms	0 - 55 MB	NPU
SINet	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	3.693 ms	1 - 4 MB	NPU
SINet	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	3.824 ms	1 - 1 MB	NPU
SINet	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	1.81 ms	1 - 33 MB	NPU
SINet	QNN_DLC	float	Qualcomm® SA7255P	6.614 ms	1 - 34 MB	NPU
SINet	QNN_DLC	float	Qualcomm® SA8295P	4.636 ms	0 - 36 MB	NPU
SINet	QNN_DLC	float	Snapdragon® 8 Elite Mobile	1.81 ms	1 - 33 MB	NPU
SINet	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	1.128 ms	0 - 36 MB	NPU
SINet	QNN_DLC	w8a16	Snapdragon® X2 Elite	3.535 ms	0 - 0 MB	NPU
SINet	QNN_DLC	w8a16	Snapdragon® X Elite	7.161 ms	0 - 0 MB	NPU
SINet	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	4.082 ms	0 - 50 MB	NPU
SINet	QNN_DLC	w8a16	Snapdragon® 8 Gen 1 Mobile	7.165 ms	0 - 55 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® Dragonwing™ QCS8275	10.825 ms	0 - 39 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	6.583 ms	0 - 32 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® SA8775P	7.009 ms	0 - 40 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® SA8650P	7.009 ms	0 - 40 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® SA8255P	7.009 ms	0 - 40 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® QCS8450	7.165 ms	0 - 55 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-9075	6.813 ms	2 - 5 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-X7181	7.161 ms	0 - 0 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-8750	3.435 ms	0 - 41 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® SA7255P	10.825 ms	0 - 39 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® SA8295P	8.135 ms	0 - 37 MB	NPU
SINet	QNN_DLC	w8a16	Snapdragon® 8 Elite Mobile	3.435 ms	0 - 41 MB	NPU
SINet	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	3.214 ms	0 - 40 MB	NPU
SINet	QNN_DLC	w8a8	Snapdragon® X2 Elite	2.331 ms	0 - 0 MB	NPU
SINet	QNN_DLC	w8a8	Snapdragon® X Elite	4.739 ms	0 - 0 MB	NPU
SINet	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	2.904 ms	0 - 47 MB	NPU
SINet	QNN_DLC	w8a8	Snapdragon® 8 Gen 1 Mobile	4.63 ms	0 - 51 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS8275	8.313 ms	0 - 36 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® Dragonwing™ QCS8550 (Proxy)	4.484 ms	0 - 2 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® SA8775P	4.679 ms	0 - 39 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® SA8650P	4.679 ms	0 - 39 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® SA8255P	4.679 ms	0 - 39 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® QCS8450	4.63 ms	0 - 51 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-9075	4.837 ms	0 - 2 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® Dragonwing™ IQ-X7181	4.739 ms	0 - 0 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® Dragonwing™ Q-8750	2.364 ms	0 - 39 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® SA7255P	8.313 ms	0 - 36 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® SA8295P	5.217 ms	0 - 35 MB	NPU
SINet	QNN_DLC	w8a8	Snapdragon® 8 Elite Mobile	2.364 ms	0 - 39 MB	NPU
SINet	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	2.14 ms	0 - 38 MB	NPU
SINet	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	2.111 ms	0 - 45 MB	NPU
SINet	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	3.769 ms	0 - 55 MB	NPU
SINet	TFLITE	float	Qualcomm® Dragonwing™ QCS8275	6.682 ms	1 - 34 MB	NPU
SINet	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	3.395 ms	0 - 2 MB	NPU
SINet	TFLITE	float	Qualcomm® SA8775P	3.894 ms	0 - 36 MB	NPU
SINet	TFLITE	float	Qualcomm® SA8650P	3.894 ms	0 - 36 MB	NPU
SINet	TFLITE	float	Qualcomm® SA8255P	3.894 ms	0 - 36 MB	NPU
SINet	TFLITE	float	Qualcomm® QCS8450	3.769 ms	0 - 55 MB	NPU
SINet	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	3.692 ms	1 - 4 MB	NPU
SINet	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	1.809 ms	0 - 36 MB	NPU
SINet	TFLITE	float	Qualcomm® SA7255P	6.682 ms	1 - 34 MB	NPU
SINet	TFLITE	float	Qualcomm® SA8295P	4.723 ms	0 - 34 MB	NPU
SINet	TFLITE	float	Snapdragon® 8 Elite Mobile	1.809 ms	0 - 36 MB	NPU
SINet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1.138 ms	0 - 34 MB	NPU

License

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

References

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.

Downloads last month: -; Downloads are not tracked for this model. How to track

Paper for qualcomm/SINet

SINet: Extreme Lightweight Portrait Segmentation Networks with Spatial Squeeze Modules and Information Blocking Decoder

Paper • 1911.09099 • Published Nov 20, 2019