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 2.37, ONNX Runtime 1.23.0	Download
ONNX	w8a8	Universal	QAIRT 2.37, ONNX Runtime 1.23.0	Download
QNN_DLC	float	Universal	QAIRT 2.42	Download
QNN_DLC	w8a16	Universal	QAIRT 2.42	Download
QNN_DLC	w8a8	Universal	QAIRT 2.42	Download
TFLITE	float	Universal	QAIRT 2.42, TFLite 2.17.0	Download
TFLITE	w8a8	Universal	QAIRT 2.42, TFLite 2.17.0	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® X Elite	1.625 ms	2 - 2 MB	NPU
SINet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.056 ms	0 - 116 MB	NPU
SINet	ONNX	float	Qualcomm® QCS8550 (Proxy)	1.632 ms	1 - 123 MB	NPU
SINet	ONNX	float	Qualcomm® QCS9075	2.048 ms	1 - 3 MB	NPU
SINet	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	0.824 ms	0 - 103 MB	NPU
SINet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.737 ms	0 - 104 MB	NPU
SINet	ONNX	w8a8	Snapdragon® X Elite	6.496 ms	6 - 6 MB	NPU
SINet	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	3.52 ms	0 - 122 MB	NPU
SINet	ONNX	w8a8	Qualcomm® QCS6490	37.434 ms	7 - 10 MB	CPU
SINet	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	4.746 ms	5 - 8 MB	NPU
SINet	ONNX	w8a8	Qualcomm® QCS9075	6.79 ms	6 - 9 MB	NPU
SINet	ONNX	w8a8	Qualcomm® QCM6690	12.6 ms	7 - 16 MB	CPU
SINet	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	2.966 ms	0 - 103 MB	NPU
SINet	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	8.576 ms	7 - 16 MB	CPU
SINet	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	2.814 ms	0 - 107 MB	NPU
SINet	QNN_DLC	float	Snapdragon® X Elite	1.884 ms	1 - 1 MB	NPU
SINet	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1.109 ms	0 - 47 MB	NPU
SINet	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	1.632 ms	1 - 2 MB	NPU
SINet	QNN_DLC	float	Qualcomm® SA8775P	8.556 ms	1 - 37 MB	NPU
SINet	QNN_DLC	float	Qualcomm® QCS9075	1.97 ms	1 - 3 MB	NPU
SINet	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	2.54 ms	0 - 52 MB	NPU
SINet	QNN_DLC	float	Qualcomm® SA8295P	2.484 ms	0 - 36 MB	NPU
SINet	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	0.795 ms	0 - 32 MB	NPU
SINet	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.645 ms	1 - 38 MB	NPU
SINet	QNN_DLC	w8a16	Snapdragon® X Elite	2.042 ms	0 - 0 MB	NPU
SINet	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	1.221 ms	0 - 49 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	3.528 ms	0 - 36 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	1.797 ms	0 - 2 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® SA8775P	2.135 ms	0 - 39 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® QCS9075	2.036 ms	2 - 4 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® QCS8450 (Proxy)	2.247 ms	0 - 51 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® SA7255P	3.528 ms	0 - 36 MB	NPU
SINet	QNN_DLC	w8a16	Qualcomm® SA8295P	2.868 ms	0 - 35 MB	NPU
SINet	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	0.913 ms	0 - 37 MB	NPU
SINet	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	0.743 ms	0 - 39 MB	NPU
SINet	QNN_DLC	w8a8	Snapdragon® X Elite	1.492 ms	0 - 0 MB	NPU
SINet	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	0.893 ms	0 - 46 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	2.549 ms	0 - 35 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	1.283 ms	0 - 2 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® SA8775P	1.534 ms	0 - 37 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® QCS9075	1.454 ms	2 - 4 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	1.463 ms	0 - 47 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® SA7255P	2.549 ms	0 - 35 MB	NPU
SINet	QNN_DLC	w8a8	Qualcomm® SA8295P	1.985 ms	0 - 33 MB	NPU
SINet	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.648 ms	0 - 38 MB	NPU
SINet	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.52 ms	0 - 37 MB	NPU
SINet	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1.087 ms	0 - 46 MB	NPU
SINet	TFLITE	float	Qualcomm® QCS8275 (Proxy)	3.603 ms	0 - 36 MB	NPU
SINet	TFLITE	float	Qualcomm® QCS8550 (Proxy)	1.635 ms	0 - 2 MB	NPU
SINet	TFLITE	float	Qualcomm® SA8775P	2.066 ms	0 - 40 MB	NPU
SINet	TFLITE	float	Qualcomm® QCS9075	1.964 ms	0 - 3 MB	NPU
SINet	TFLITE	float	Qualcomm® QCS8450 (Proxy)	2.552 ms	0 - 50 MB	NPU
SINet	TFLITE	float	Qualcomm® SA7255P	3.603 ms	0 - 36 MB	NPU
SINet	TFLITE	float	Qualcomm® SA8295P	2.488 ms	0 - 36 MB	NPU
SINet	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	0.8 ms	0 - 38 MB	NPU
SINet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	0.638 ms	0 - 38 MB	NPU
SINet	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	0.843 ms	0 - 44 MB	NPU
SINet	TFLITE	w8a8	Qualcomm® QCS6490	15.492 ms	0 - 12 MB	NPU
SINet	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	28.29 ms	1 - 19 MB	CPU
SINet	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	1.215 ms	0 - 2 MB	NPU
SINet	TFLITE	w8a8	Qualcomm® SA8775P	1.489 ms	0 - 38 MB	NPU
SINet	TFLITE	w8a8	Qualcomm® QCS9075	1.373 ms	0 - 3 MB	NPU
SINet	TFLITE	w8a8	Qualcomm® QCM6690	14.238 ms	0 - 31 MB	NPU
SINet	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	1.415 ms	0 - 44 MB	NPU
SINet	TFLITE	w8a8	Qualcomm® SA7255P	28.29 ms	1 - 19 MB	CPU
SINet	TFLITE	w8a8	Qualcomm® SA8295P	1.919 ms	0 - 32 MB	NPU
SINet	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.631 ms	0 - 38 MB	NPU
SINet	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	5.701 ms	0 - 35 MB	NPU
SINet	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.554 ms	0 - 36 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