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

---

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

# Unet-Segmentation: Optimized for Qualcomm Devices

UNet is a machine learning model that produces a segmentation mask for an image. The most basic use case will label each pixel in the image as being in the foreground or the background. More advanced usage will assign a class label to each pixel. This version of the model was trained on the data from Kaggle's Carvana Image Masking Challenge (see https://www.kaggle.com/c/carvana-image-masking-challenge) and is used for vehicle segmentation.

This is based on the implementation of Unet-Segmentation found [here](https://github.com/milesial/Pytorch-UNet).
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/v0.58.0/src/qai_hub_models/models/unet_segmentation) 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/unet_segmentation/releases/v0.58.0/unet_segmentation-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/unet_segmentation/releases/v0.58.0/unet_segmentation-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/unet_segmentation/releases/v0.58.0/unet_segmentation-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/unet_segmentation/releases/v0.58.0/unet_segmentation-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/unet_segmentation/releases/v0.58.0/unet_segmentation-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/unet_segmentation/releases/v0.58.0/unet_segmentation-tflite-w8a8.zip)

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


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/unet_segmentation) 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 [Unet-Segmentation on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/unet_segmentation) for usage instructions.

## Model Details

**Model Type:** Model_use_case.semantic_segmentation

**Model Stats:**
- Model checkpoint: unet_carvana_scale1.0_epoch2
- Input resolution: 640x1280
- Number of output classes: 2 (foreground / background)
- Number of parameters: 31.0M
- Model size (float): 118 MB
- Model size (w8a8): 29.8 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| Unet-Segmentation | ONNX | float | Snapdragon® X2 Elite | 74.758 ms | 17 - 17 MB | NPU
| Unet-Segmentation | ONNX | float | Snapdragon® X Elite | 142.19 ms | 54 - 54 MB | NPU
| Unet-Segmentation | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 112.583 ms | 3 - 525 MB | NPU
| Unet-Segmentation | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 280.746 ms | 23 - 573 MB | NPU
| Unet-Segmentation | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 149.053 ms | 0 - 58 MB | NPU
| Unet-Segmentation | ONNX | float | Qualcomm® QCS8450 | 280.746 ms | 23 - 573 MB | NPU
| Unet-Segmentation | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 250.238 ms | 9 - 21 MB | NPU
| Unet-Segmentation | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 65.791 ms | 15 - 343 MB | NPU
| Unet-Segmentation | ONNX | float | Snapdragon® 8 Elite Mobile | 91.148 ms | 15 - 333 MB | NPU
| Unet-Segmentation | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 91.148 ms | 15 - 333 MB | NPU
| Unet-Segmentation | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 142.19 ms | 54 - 54 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Snapdragon® X2 Elite | 18.765 ms | 5 - 5 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Snapdragon® X Elite | 37.742 ms | 29 - 29 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 29.539 ms | 6 - 339 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 67.125 ms | 6 - 341 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 299.926 ms | 3 - 8 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 38.164 ms | 0 - 44 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Qualcomm® QCS8450 | 67.125 ms | 6 - 341 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 24.427 ms | 3 - 189 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 81.755 ms | 6 - 283 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 16.664 ms | 3 - 192 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 35.632 ms | 4 - 7 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 1212.612 ms | 0 - 540 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 81.755 ms | 6 - 283 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 24.427 ms | 3 - 189 MB | NPU
| Unet-Segmentation | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 37.742 ms | 29 - 29 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Snapdragon® X2 Elite | 71.814 ms | 9 - 9 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Snapdragon® X Elite | 132.427 ms | 9 - 9 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 102.328 ms | 9 - 523 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 269.147 ms | 4 - 538 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Qualcomm® QCS8275 | 953.606 ms | 1 - 323 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 136.602 ms | 10 - 12 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Qualcomm® SA8775P | 240.481 ms | 0 - 323 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Qualcomm® SA8650P | 240.481 ms | 0 - 323 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Qualcomm® SA8255P | 240.481 ms | 0 - 323 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Qualcomm® QCS8450 | 269.147 ms | 4 - 538 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 248.054 ms | 9 - 27 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 62.699 ms | 9 - 355 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Qualcomm® SA7255P | 953.606 ms | 1 - 323 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 82.425 ms | 9 - 341 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Qualcomm® SA8295P | 274.443 ms | 0 - 322 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 82.425 ms | 9 - 341 MB | NPU
| Unet-Segmentation | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 132.427 ms | 9 - 9 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 18.85 ms | 2 - 2 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Snapdragon® X Elite | 35.759 ms | 2 - 2 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 26.158 ms | 2 - 319 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 59.182 ms | 2 - 317 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 289.343 ms | 2 - 8 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® QCS8275 | 121.495 ms | 1 - 179 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 34.862 ms | 2 - 4 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® SA8775P | 32.168 ms | 1 - 180 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® SA8650P | 32.168 ms | 1 - 180 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® SA8255P | 32.168 ms | 1 - 180 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 59.182 ms | 2 - 317 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 21.797 ms | 2 - 190 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 78.995 ms | 2 - 272 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 16.047 ms | 2 - 201 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 32.518 ms | 1 - 7 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 1224.76 ms | 3 - 524 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® SA7255P | 121.495 ms | 1 - 179 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® SA8295P | 63.754 ms | 0 - 179 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 78.995 ms | 2 - 272 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 21.797 ms | 2 - 190 MB | NPU
| Unet-Segmentation | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 35.759 ms | 2 - 2 MB | NPU
| Unet-Segmentation | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 102.101 ms | 5 - 576 MB | NPU
| Unet-Segmentation | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 276.402 ms | 7 - 589 MB | NPU
| Unet-Segmentation | TFLITE | float | Qualcomm® QCS8275 | 953.406 ms | 0 - 324 MB | NPU
| Unet-Segmentation | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 136.461 ms | 6 - 219 MB | NPU
| Unet-Segmentation | TFLITE | float | Qualcomm® SA8775P | 240.517 ms | 7 - 330 MB | NPU
| Unet-Segmentation | TFLITE | float | Qualcomm® SA8650P | 240.517 ms | 7 - 330 MB | NPU
| Unet-Segmentation | TFLITE | float | Qualcomm® SA8255P | 240.517 ms | 7 - 330 MB | NPU
| Unet-Segmentation | TFLITE | float | Qualcomm® QCS8450 | 276.402 ms | 7 - 589 MB | NPU
| Unet-Segmentation | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 244.108 ms | 0 - 80 MB | NPU
| Unet-Segmentation | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 61.29 ms | 6 - 353 MB | NPU
| Unet-Segmentation | TFLITE | float | Qualcomm® SA7255P | 953.406 ms | 0 - 324 MB | NPU
| Unet-Segmentation | TFLITE | float | Snapdragon® 8 Elite Mobile | 82.365 ms | 0 - 331 MB | NPU
| Unet-Segmentation | TFLITE | float | Qualcomm® SA8295P | 274.471 ms | 7 - 328 MB | NPU
| Unet-Segmentation | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 82.365 ms | 0 - 331 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 26.157 ms | 1 - 318 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 60.638 ms | 2 - 318 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 288.989 ms | 1 - 41 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® QCS8275 | 121.583 ms | 2 - 180 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 34.611 ms | 2 - 4 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® SA8775P | 32.212 ms | 2 - 180 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® SA8650P | 32.212 ms | 2 - 180 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® SA8255P | 32.212 ms | 2 - 180 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® QCS8450 | 60.638 ms | 2 - 318 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 22.028 ms | 2 - 188 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 78.944 ms | 1 - 264 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 16.078 ms | 2 - 199 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 32.629 ms | 1 - 38 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 1231.001 ms | 1 - 522 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® SA7255P | 121.583 ms | 2 - 180 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® SA8295P | 63.758 ms | 2 - 180 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 78.944 ms | 1 - 264 MB | NPU
| Unet-Segmentation | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 22.028 ms | 2 - 188 MB | NPU

## License
* The license for the original implementation of Unet-Segmentation can be found
  [here](https://github.com/milesial/Pytorch-UNet/blob/master/LICENSE).

## References
* [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://arxiv.org/abs/1505.04597)
* [Source Model Implementation](https://github.com/milesial/Pytorch-UNet)

## 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).