---
library_name: pytorch
license: other
tags:
- android
pipeline_tag: image-segmentation

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sinet/web-assets/model_demo.png)

# 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](https://github.com/clovaai/ext_portrait_segmentation).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/sinet) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).

Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) 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](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sinet/releases/v0.51.0/sinet-onnx-float.zip)
| ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sinet/releases/v0.51.0/sinet-onnx-w8a16.zip)
| ONNX | w8a8 | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sinet/releases/v0.51.0/sinet-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sinet/releases/v0.51.0/sinet-qnn_dlc-float.zip)
| QNN_DLC | w8a16 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sinet/releases/v0.51.0/sinet-qnn_dlc-w8a16.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sinet/releases/v0.51.0/sinet-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sinet/releases/v0.51.0/sinet-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sinet/releases/v0.51.0/sinet-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[SINet on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/sinet)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/sinet) 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](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/sinet) 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® 8 Elite Gen 5 Mobile | 0.708 ms | 0 - 40 MB | NPU
| SINet | ONNX | float | Snapdragon® X2 Elite | 0.769 ms | 0 - 0 MB | NPU
| SINet | ONNX | float | Snapdragon® X Elite | 1.935 ms | 2 - 2 MB | NPU
| SINet | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 1.097 ms | 0 - 46 MB | NPU
| SINet | ONNX | float | Qualcomm® QCS8550 (Proxy) | 1.685 ms | 0 - 2 MB | NPU
| SINet | ONNX | float | Qualcomm® QCS9075 | 2.142 ms | 2 - 5 MB | NPU
| SINet | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.853 ms | 0 - 29 MB | NPU
| SINet | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.887 ms | 0 - 45 MB | NPU
| SINet | ONNX | w8a16 | Snapdragon® X2 Elite | 0.958 ms | 0 - 0 MB | NPU
| SINet | ONNX | w8a16 | Snapdragon® X Elite | 2.247 ms | 1 - 1 MB | NPU
| SINet | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1.309 ms | 0 - 53 MB | NPU
| SINet | ONNX | w8a16 | Qualcomm® QCS6490 | 53.8 ms | 14 - 17 MB | CPU
| SINet | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 1.984 ms | 0 - 3 MB | NPU
| SINet | ONNX | w8a16 | Qualcomm® QCS9075 | 2.249 ms | 0 - 3 MB | NPU
| SINet | ONNX | w8a16 | Qualcomm® QCM6690 | 19.606 ms | 22 - 29 MB | CPU
| SINet | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 1.036 ms | 0 - 38 MB | NPU
| SINet | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 13.779 ms | 23 - 31 MB | CPU
| SINet | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 3.356 ms | 0 - 37 MB | NPU
| SINet | ONNX | w8a8 | Snapdragon® X2 Elite | 3.342 ms | 8 - 8 MB | NPU
| SINet | ONNX | w8a8 | Snapdragon® X Elite | 44.017 ms | 7 - 7 MB | NPU
| SINet | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 4.559 ms | 7 - 61 MB | NPU
| SINet | ONNX | w8a8 | Qualcomm® QCS6490 | 37.882 ms | 7 - 10 MB | CPU
| SINet | ONNX | w8a8 | Qualcomm® QCM6690 | 16.09 ms | 7 - 16 MB | CPU
| SINet | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 3.468 ms | 0 - 42 MB | NPU
| SINet | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 11.247 ms | 7 - 15 MB | CPU
| SINet | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.126 ms | 1 - 38 MB | NPU
| SINet | QNN_DLC | float | Snapdragon® X2 Elite | 1.44 ms | 1 - 1 MB | NPU
| SINet | QNN_DLC | float | Snapdragon® X Elite | 3.805 ms | 1 - 1 MB | NPU
| SINet | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 2.101 ms | 0 - 46 MB | NPU
| SINet | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 6.623 ms | 1 - 35 MB | NPU
| SINet | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 3.43 ms | 1 - 3 MB | NPU
| SINet | QNN_DLC | float | Qualcomm® SA8775P | 3.893 ms | 1 - 36 MB | NPU
| SINet | QNN_DLC | float | Qualcomm® QCS9075 | 3.83 ms | 1 - 4 MB | NPU
| SINet | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 3.746 ms | 0 - 51 MB | NPU
| SINet | QNN_DLC | float | Qualcomm® SA7255P | 6.623 ms | 1 - 35 MB | NPU
| SINet | QNN_DLC | float | Qualcomm® SA8295P | 4.657 ms | 0 - 35 MB | NPU
| SINet | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.812 ms | 1 - 37 MB | NPU
| SINet | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 3.205 ms | 0 - 42 MB | NPU
| SINet | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 3.543 ms | 0 - 0 MB | NPU
| SINet | QNN_DLC | w8a16 | Snapdragon® X Elite | 7.119 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 | Qualcomm® QCS8275 (Proxy) | 10.797 ms | 0 - 37 MB | NPU
| SINet | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 6.55 ms | 0 - 2 MB | NPU
| SINet | QNN_DLC | w8a16 | Qualcomm® SA8775P | 7.057 ms | 0 - 41 MB | NPU
| SINet | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 6.851 ms | 2 - 4 MB | NPU
| SINet | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 7.187 ms | 0 - 52 MB | NPU
| SINet | QNN_DLC | w8a16 | Qualcomm® SA7255P | 10.797 ms | 0 - 37 MB | NPU
| SINet | QNN_DLC | w8a16 | Qualcomm® SA8295P | 8.183 ms | 0 - 37 MB | NPU
| SINet | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 3.437 ms | 0 - 41 MB | NPU
| SINet | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 2.135 ms | 0 - 40 MB | NPU
| SINet | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 2.353 ms | 0 - 0 MB | NPU
| SINet | QNN_DLC | w8a8 | Snapdragon® X Elite | 4.9 ms | 0 - 0 MB | NPU
| SINet | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 3.006 ms | 0 - 47 MB | NPU
| SINet | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 8.015 ms | 0 - 35 MB | NPU
| SINet | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 4.5 ms | 0 - 10 MB | NPU
| SINet | QNN_DLC | w8a8 | Qualcomm® SA8775P | 4.691 ms | 0 - 39 MB | NPU
| SINet | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 4.438 ms | 2 - 4 MB | NPU
| SINet | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 4.633 ms | 0 - 49 MB | NPU
| SINet | QNN_DLC | w8a8 | Qualcomm® SA7255P | 8.015 ms | 0 - 35 MB | NPU
| SINet | QNN_DLC | w8a8 | Qualcomm® SA8295P | 5.245 ms | 0 - 35 MB | NPU
| SINet | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 2.354 ms | 0 - 40 MB | NPU
| SINet | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.687 ms | 0 - 38 MB | NPU
| SINet | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.077 ms | 0 - 44 MB | NPU
| SINet | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 3.749 ms | 1 - 33 MB | NPU
| SINet | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 1.628 ms | 0 - 2 MB | NPU
| SINet | TFLITE | float | Qualcomm® SA8775P | 2.101 ms | 0 - 35 MB | NPU
| SINet | TFLITE | float | Qualcomm® QCS9075 | 2.076 ms | 1 - 4 MB | NPU
| SINet | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 2.024 ms | 0 - 49 MB | NPU
| SINet | TFLITE | float | Qualcomm® SA7255P | 3.749 ms | 1 - 33 MB | NPU
| SINet | TFLITE | float | Qualcomm® SA8295P | 2.502 ms | 0 - 33 MB | NPU
| SINet | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 0.817 ms | 0 - 34 MB | NPU
| SINet | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.632 ms | 0 - 40 MB | NPU
| SINet | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 1.023 ms | 0 - 46 MB | NPU
| SINet | TFLITE | w8a8 | Qualcomm® QCS6490 | 22.698 ms | 0 - 13 MB | NPU
| SINet | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 2.931 ms | 0 - 35 MB | NPU
| SINet | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 1.477 ms | 0 - 2 MB | NPU
| SINet | TFLITE | w8a8 | Qualcomm® SA8775P | 1.772 ms | 0 - 39 MB | NPU
| SINet | TFLITE | w8a8 | Qualcomm® QCS9075 | 1.767 ms | 0 - 3 MB | NPU
| SINet | TFLITE | w8a8 | Qualcomm® QCM6690 | 12.81 ms | 0 - 29 MB | NPU
| SINet | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 1.718 ms | 0 - 46 MB | NPU
| SINet | TFLITE | w8a8 | Qualcomm® SA7255P | 2.931 ms | 0 - 35 MB | NPU
| SINet | TFLITE | w8a8 | Qualcomm® SA8295P | 2.238 ms | 0 - 35 MB | NPU
| SINet | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.745 ms | 0 - 33 MB | NPU
| SINet | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 6.049 ms | 0 - 30 MB | NPU

## License
* The license for the original implementation of SINet can be found
  [here](https://github.com/clovaai/ext_portrait_segmentation/blob/master/LICENSE).

## References
* [SINet: Extreme Lightweight Portrait Segmentation Networks with Spatial Squeeze Modules and Information Blocking Decoder](https://arxiv.org/abs/1911.09099)
* [Source Model Implementation](https://github.com/clovaai/ext_portrait_segmentation)

## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).