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 Mobile Deployment
## Segment images or video by class in real-time on device


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

This model is an implementation of PidNet found [here](https://github.com/XuJiacong/PIDNet).


This repository provides scripts to run PidNet on Qualcomm® devices.
More details on model performance across various devices, can be found
[here](https://aihub.qualcomm.com/models/pidnet).



### 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

| Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model
|---|---|---|---|---|---|---|---|---|
| PidNet | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 115.583 ms | 0 - 194 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.tflite) |
| PidNet | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 115.619 ms | 21 - 207 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.dlc) |
| PidNet | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 69.254 ms | 2 - 305 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.tflite) |
| PidNet | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 68.751 ms | 21 - 314 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.dlc) |
| PidNet | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 35.818 ms | 2 - 5 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.tflite) |
| PidNet | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 35.971 ms | 24 - 26 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.dlc) |
| PidNet | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 29.876 ms | 24 - 47 MB | NPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.onnx.zip) |
| PidNet | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 201.947 ms | 0 - 195 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.tflite) |
| PidNet | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 44.451 ms | 24 - 209 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.dlc) |
| PidNet | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 115.583 ms | 0 - 194 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.tflite) |
| PidNet | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 115.619 ms | 21 - 207 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.dlc) |
| PidNet | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 50.417 ms | 2 - 208 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.tflite) |
| PidNet | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 50.403 ms | 24 - 222 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.dlc) |
| PidNet | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 201.947 ms | 0 - 195 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.tflite) |
| PidNet | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 44.451 ms | 24 - 209 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.dlc) |
| PidNet | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 24.276 ms | 1 - 301 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.tflite) |
| PidNet | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 24.351 ms | 24 - 307 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.dlc) |
| PidNet | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 19.345 ms | 31 - 282 MB | NPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.onnx.zip) |
| PidNet | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 18.53 ms | 2 - 211 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.tflite) |
| PidNet | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 18.553 ms | 22 - 228 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.dlc) |
| PidNet | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 16.225 ms | 7 - 168 MB | NPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.onnx.zip) |
| PidNet | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | TFLITE | 13.623 ms | 2 - 229 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.tflite) |
| PidNet | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_DLC | 13.814 ms | 24 - 247 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.dlc) |
| PidNet | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | ONNX | 11.491 ms | 30 - 248 MB | NPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.onnx.zip) |
| PidNet | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 36.89 ms | 24 - 24 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.dlc) |
| PidNet | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 29.757 ms | 24 - 24 MB | NPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet.onnx.zip) |
| PidNet | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® QCM6690 | TFLITE | 220.377 ms | 2 - 172 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® QCM6690 | ONNX | 352.863 ms | 178 - 195 MB | CPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.onnx.zip) |
| PidNet | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | TFLITE | 184.205 ms | 2 - 72 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | ONNX | 391.806 ms | 195 - 216 MB | CPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.onnx.zip) |
| PidNet | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 132.43 ms | 1 - 180 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 125.083 ms | 6 - 186 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.dlc) |
| PidNet | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 74.938 ms | 1 - 235 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 77.218 ms | 6 - 240 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.dlc) |
| PidNet | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 52.739 ms | 1 - 3 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 64.927 ms | 6 - 8 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.dlc) |
| PidNet | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 60.761 ms | 99 - 101 MB | NPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.onnx.zip) |
| PidNet | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 53.578 ms | 1 - 180 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 65.762 ms | 6 - 186 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.dlc) |
| PidNet | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 132.43 ms | 1 - 180 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 125.083 ms | 6 - 186 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.dlc) |
| PidNet | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 76.863 ms | 1 - 183 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 73.627 ms | 6 - 188 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.dlc) |
| PidNet | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 53.578 ms | 1 - 180 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 65.762 ms | 6 - 186 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.dlc) |
| PidNet | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 51.133 ms | 1 - 238 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 48.374 ms | 6 - 239 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.dlc) |
| PidNet | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 46.624 ms | 105 - 315 MB | NPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.onnx.zip) |
| PidNet | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 37.933 ms | 1 - 197 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 41.018 ms | 6 - 204 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.dlc) |
| PidNet | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 42.002 ms | 98 - 254 MB | NPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.onnx.zip) |
| PidNet | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | TFLITE | 86.82 ms | 2 - 192 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | ONNX | 327.892 ms | 190 - 207 MB | CPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.onnx.zip) |
| PidNet | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | TFLITE | 36.7 ms | 3 - 224 MB | NPU | [PidNet.tflite](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.tflite) |
| PidNet | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_DLC | 43.109 ms | 6 - 226 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.dlc) |
| PidNet | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | ONNX | 40.896 ms | 106 - 271 MB | NPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.onnx.zip) |
| PidNet | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 68.011 ms | 6 - 6 MB | NPU | [PidNet.dlc](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.dlc) |
| PidNet | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 62.67 ms | 131 - 131 MB | NPU | [PidNet.onnx.zip](https://huggingface.co/qualcomm/PidNet/blob/main/PidNet_w8a8.onnx.zip) |




## Installation


Install the package via pip:
```bash
pip install qai-hub-models
```


## Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device

Sign-in to [Qualcomm® AI Hub Workbench](https://workbench.aihub.qualcomm.com/) with your
Qualcomm® ID. Once signed in navigate to `Account -> Settings -> API Token`.

With this API token, you can configure your client to run models on the cloud
hosted devices.
```bash
qai-hub configure --api_token API_TOKEN
```
Navigate to [docs](https://workbench.aihub.qualcomm.com/docs/) for more information.



## Demo off target

The package contains a simple end-to-end demo that downloads pre-trained
weights and runs this model on a sample input.

```bash
python -m qai_hub_models.models.pidnet.demo
```

The above demo runs a reference implementation of pre-processing, model
inference, and post processing.

**NOTE**: If you want running in a Jupyter Notebook or Google Colab like
environment, please add the following to your cell (instead of the above).
```
%run -m qai_hub_models.models.pidnet.demo
```


### Run model on a cloud-hosted device

In addition to the demo, you can also run the model on a cloud-hosted Qualcomm®
device. This script does the following:
* Performance check on-device on a cloud-hosted device
* Downloads compiled assets that can be deployed on-device for Android.
* Accuracy check between PyTorch and on-device outputs.

```bash
python -m qai_hub_models.models.pidnet.export
```



## How does this work?

This [export script](https://aihub.qualcomm.com/models/pidnet/qai_hub_models/models/PidNet/export.py)
leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) to optimize, validate, and deploy this model
on-device. Lets go through each step below in detail:

Step 1: **Compile model for on-device deployment**

To compile a PyTorch model for on-device deployment, we first trace the model
in memory using the `jit.trace` and then call the `submit_compile_job` API.

```python
import torch

import qai_hub as hub
from qai_hub_models.models.pidnet import Model

# Load the model
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S25")

# Trace model
input_shape = torch_model.get_input_spec()
sample_inputs = torch_model.sample_inputs()

pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])

# Compile model on a specific device
compile_job = hub.submit_compile_job(
    model=pt_model,
    device=device,
    input_specs=torch_model.get_input_spec(),
)

# Get target model to run on-device
target_model = compile_job.get_target_model()

```


Step 2: **Performance profiling on cloud-hosted device**

After compiling models from step 1. Models can be profiled model on-device using the
`target_model`. Note that this scripts runs the model on a device automatically
provisioned in the cloud.  Once the job is submitted, you can navigate to a
provided job URL to view a variety of on-device performance metrics.
```python
profile_job = hub.submit_profile_job(
    model=target_model,
    device=device,
)
        
```

Step 3: **Verify on-device accuracy**

To verify the accuracy of the model on-device, you can run on-device inference
on sample input data on the same cloud hosted device.
```python
input_data = torch_model.sample_inputs()
inference_job = hub.submit_inference_job(
    model=target_model,
    device=device,
    inputs=input_data,
)
    on_device_output = inference_job.download_output_data()

```
With the output of the model, you can compute like PSNR, relative errors or
spot check the output with expected output.

**Note**: This on-device profiling and inference requires access to Qualcomm®
AI Hub Workbench. [Sign up for access](https://myaccount.qualcomm.com/signup).



## Run demo on a cloud-hosted device

You can also run the demo on-device.

```bash
python -m qai_hub_models.models.pidnet.demo --eval-mode on-device
```

**NOTE**: If you want running in a Jupyter Notebook or Google Colab like
environment, please add the following to your cell (instead of the above).
```
%run -m qai_hub_models.models.pidnet.demo -- --eval-mode on-device
```


## Deploying compiled model to Android


The models can be deployed using multiple runtimes:
- TensorFlow Lite (`.tflite` export): [This
  tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a
  guide to deploy the .tflite model in an Android application.


- QNN (`.so` export ): This [sample
  app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html)
provides instructions on how to use the `.so` shared library  in an Android application.


## View on Qualcomm® AI Hub
Get more details on PidNet's performance across various devices [here](https://aihub.qualcomm.com/models/pidnet).
Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)


## 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).