README.md

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

---

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

# PidNet: Optimized for Qualcomm Devices

PIDNet (Proportional-Integral-Derivative Network) is a real-time semantic segmentation model based on  PID controllers

This is based on the implementation of PidNet found [here](https://github.com/XuJiacong/PIDNet).
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/v0.57.3/src/qai_hub_models/models/pidnet) 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 2.45, ONNX Runtime 1.25.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pidnet/releases/v0.57.3/pidnet-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.45, ONNX Runtime 1.25.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pidnet/releases/v0.57.3/pidnet-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/pidnet/releases/v0.57.3/pidnet-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pidnet/releases/v0.57.3/pidnet-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/pidnet/releases/v0.57.3/pidnet-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pidnet/releases/v0.57.3/pidnet-tflite-w8a8.zip)

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


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.57.3/src/qai_hub_models/models/pidnet) 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 [PidNet on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.57.3/src/qai_hub_models/models/pidnet) for usage instructions.

## Model Details

**Model Type:** Model_use_case.semantic_segmentation

**Model Stats:**
- Model checkpoint: PIDNet_S_Cityscapes_val.pt
- Inference latency: RealTime
- Input resolution: 1024x2048
- Number of output classes: 19
- Number of parameters: 8.06M
- Model size (float): 29.1 MB
- Model size (w8a8): 8.02 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| PidNet | ONNX | float | Snapdragon® X2 Elite | 13.763 ms | 189 - 189 MB | NPU
| PidNet | ONNX | float | Snapdragon® X Elite | 32.32 ms | 158 - 158 MB | NPU
| PidNet | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 22.916 ms | 31 - 322 MB | NPU
| PidNet | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 55.626 ms | 29 - 324 MB | NPU
| PidNet | ONNX | float | Qualcomm® QCS8550 (Proxy) | 32.415 ms | 24 - 51 MB | NPU
| PidNet | ONNX | float | Qualcomm® QCS8450 | 55.626 ms | 29 - 324 MB | NPU
| PidNet | ONNX | float | Snapdragon® 8 Elite Mobile | 17.854 ms | 7 - 219 MB | NPU
| PidNet | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 13.144 ms | 5 - 250 MB | NPU
| PidNet | ONNX | float | Qualcomm® QCS9075 | 47.664 ms | 24 - 93 MB | NPU
| PidNet | ONNX | float | Qualcomm® QCS8750 | 17.854 ms | 7 - 219 MB | NPU
| PidNet | ONNX | float | Qualcomm® QCS7181 | 32.32 ms | 158 - 158 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® X2 Elite | 25.958 ms | 207 - 207 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® X Elite | 54.825 ms | 175 - 175 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 39.101 ms | 7 - 261 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 57.664 ms | 7 - 263 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® QCS6490 | 132.062 ms | 6 - 51 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 52.469 ms | 0 - 19 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® QCS8450 | 57.664 ms | 7 - 263 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 25.794 ms | 2 - 232 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® QCM6690 | 215.567 ms | 7 - 222 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® QCS9075 | 54.995 ms | 4 - 51 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 45.245 ms | 2 - 219 MB | NPU
| PidNet | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 61.984 ms | 7 - 232 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® QCS7790 | 61.984 ms | 7 - 232 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® QCS8750 | 45.245 ms | 2 - 219 MB | NPU
| PidNet | ONNX | w8a8 | Qualcomm® QCS7181 | 54.825 ms | 175 - 175 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® X2 Elite | 13.6 ms | 24 - 24 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® X Elite | 39.147 ms | 24 - 24 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 25.783 ms | 23 - 327 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 75.878 ms | 24 - 331 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® QCS8275 | 117.745 ms | 24 - 243 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 37.694 ms | 24 - 26 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® QCS8450 | 75.878 ms | 24 - 331 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 18.292 ms | 23 - 269 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® SA8295P | 52.014 ms | 24 - 252 MB | NPU
| PidNet | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 11.969 ms | 18 - 291 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® SA7255P | 117.745 ms | 24 - 243 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® QCS9075 | 61.104 ms | 26 - 54 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® QCS8750 | 18.292 ms | 23 - 269 MB | NPU
| PidNet | QNN_DLC | float | Qualcomm® QCS7181 | 39.147 ms | 24 - 24 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 25.202 ms | 6 - 6 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® X Elite | 60.613 ms | 6 - 6 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 43.114 ms | 6 - 266 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 63.839 ms | 6 - 269 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® QCS8275 | 111.317 ms | 6 - 219 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 57.755 ms | 6 - 158 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 63.839 ms | 6 - 269 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 24.591 ms | 6 - 273 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 61.722 ms | 6 - 14 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® SA7255P | 111.317 ms | 6 - 219 MB | NPU
| PidNet | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 43.837 ms | 6 - 244 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® SA8295P | 66.422 ms | 6 - 223 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® QCS8750 | 43.837 ms | 6 - 244 MB | NPU
| PidNet | QNN_DLC | w8a8 | Qualcomm® QCS7181 | 60.613 ms | 6 - 6 MB | NPU
| PidNet | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 24.938 ms | 2 - 325 MB | NPU
| PidNet | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 73.971 ms | 4 - 330 MB | NPU
| PidNet | TFLITE | float | Qualcomm® QCS8275 | 116.592 ms | 3 - 232 MB | NPU
| PidNet | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 36.097 ms | 2 - 172 MB | NPU
| PidNet | TFLITE | float | Qualcomm® SA8775P | 141.849 ms | 3 - 32 MB | GPU
| PidNet | TFLITE | float | Qualcomm® SA8650P | 141.849 ms | 3 - 32 MB | GPU
| PidNet | TFLITE | float | Qualcomm® SA8255P | 141.849 ms | 3 - 32 MB | GPU
| PidNet | TFLITE | float | Qualcomm® QCS8450 | 73.971 ms | 4 - 330 MB | NPU
| PidNet | TFLITE | float | Snapdragon® 8 Elite Mobile | 18.758 ms | 2 - 252 MB | NPU
| PidNet | TFLITE | float | Qualcomm® SA8295P | 51.03 ms | 3 - 245 MB | NPU
| PidNet | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 12.676 ms | 2 - 273 MB | NPU
| PidNet | TFLITE | float | Qualcomm® SA7255P | 116.592 ms | 3 - 232 MB | NPU
| PidNet | TFLITE | float | Qualcomm® QCS9075 | 59.506 ms | 0 - 45 MB | NPU
| PidNet | TFLITE | float | Qualcomm® QCS8750 | 18.758 ms | 2 - 252 MB | NPU
| PidNet | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 37.894 ms | 1 - 265 MB | NPU
| PidNet | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 58.239 ms | 1 - 265 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® QCS6490 | 185.168 ms | 2 - 71 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® QCS8275 | 98.412 ms | 1 - 214 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 50.392 ms | 1 - 3 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® SA8775P | 148.903 ms | 27 - 58 MB | GPU
| PidNet | TFLITE | w8a8 | Qualcomm® SA8650P | 148.903 ms | 27 - 58 MB | GPU
| PidNet | TFLITE | w8a8 | Qualcomm® SA8255P | 148.903 ms | 27 - 58 MB | GPU
| PidNet | TFLITE | w8a8 | Qualcomm® QCS8450 | 58.239 ms | 1 - 265 MB | NPU
| PidNet | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 22.081 ms | 0 - 267 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® QCM6690 | 236.112 ms | 3 - 234 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® QCS9075 | 53.51 ms | 1 - 17 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® SA7255P | 98.412 ms | 1 - 214 MB | NPU
| PidNet | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 69.966 ms | 1 - 234 MB | NPU
| PidNet | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 66.736 ms | 5 - 223 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® SA8295P | 57.998 ms | 1 - 218 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® QCS7790 | 66.736 ms | 5 - 223 MB | NPU
| PidNet | TFLITE | w8a8 | Qualcomm® QCS8750 | 69.966 ms | 1 - 234 MB | NPU

## License
* The license for the original implementation of PidNet can be found
  [here](https://github.com/XuJiacong/PIDNet/blob/main/LICENSE).

## References
* [PIDNet A Real-time Semantic Segmentation Network Inspired from PID Controller Segmentation of Road Scenes](https://arxiv.org/abs/2206.02066)
* [Source Model Implementation](https://github.com/XuJiacong/PIDNet)

## 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).