README.md

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

---

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

# PSPNet: Optimized for Qualcomm Devices

PSPNet (Pyramid Scene Parsing Network) is a semantic segmentation model that captures global context information by applying pyramid pooling modules. It is designed to improve scene understanding by aggregating contextual features at multiple scales.

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/v0.59.0/src/qai_hub_models/models/pspnet) 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 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pspnet/releases/v0.59.0/pspnet-onnx-float.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pspnet/releases/v0.59.0/pspnet-qnn_dlc-float.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pspnet/releases/v0.59.0/pspnet-tflite-float.zip)

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


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/pspnet) 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 [PSPNet on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/pspnet) for usage instructions.

## Model Details

**Model Type:** Model_use_case.semantic_segmentation

**Model Stats:**
- Model checkpoint: pspnet101_ade20k.pth
- Input resolution: 1x3x473x473
- Number of parameters: 65.7M
- Model size (float): 251 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| PSPNet | ONNX | float | Snapdragon® X2 Elite | 835.974 ms | 528 - 528 MB | NPU
| PSPNet | ONNX | float | Snapdragon® X Elite | 1334.498 ms | 267 - 267 MB | NPU
| PSPNet | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 956.615 ms | 142 - 1994 MB | NPU
| PSPNet | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 3276.923 ms | 34 - 887 MB | NPU
| PSPNet | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1164.165 ms | 0 - 160 MB | NPU
| PSPNet | ONNX | float | Qualcomm® QCS8450 | 3276.923 ms | 34 - 887 MB | NPU
| PSPNet | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 1417.954 ms | 124 - 130 MB | NPU
| PSPNet | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 1334.498 ms | 267 - 267 MB | NPU
| PSPNet | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 649.02 ms | 133 - 1599 MB | NPU
| PSPNet | ONNX | float | Snapdragon® 8 Elite Mobile | 649.02 ms | 133 - 1599 MB | NPU
| PSPNet | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 718.907 ms | 244 - 1842 MB | NPU
| PSPNet | QNN_DLC | float | Snapdragon® X2 Elite | 2497.872 ms | 3 - 3 MB | NPU
| PSPNet | QNN_DLC | float | Snapdragon® X Elite | 2536.944 ms | 3 - 3 MB | NPU
| PSPNet | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 1836.59 ms | 25 - 1690 MB | NPU
| PSPNet | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 1595.526 ms | 0 - 851 MB | NPU
| PSPNet | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 5274.515 ms | 13 - 1329 MB | NPU
| PSPNet | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2479.634 ms | 3 - 6 MB | NPU
| PSPNet | QNN_DLC | float | Qualcomm® SA8775P | 2606.587 ms | 2 - 1319 MB | NPU
| PSPNet | QNN_DLC | float | Qualcomm® SA8650P | 2606.587 ms | 2 - 1319 MB | NPU
| PSPNet | QNN_DLC | float | Qualcomm® SA8255P | 2606.587 ms | 2 - 1319 MB | NPU
| PSPNet | QNN_DLC | float | Qualcomm® QCS8450 | 1595.526 ms | 0 - 851 MB | NPU
| PSPNet | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 2593.163 ms | 3 - 135 MB | NPU
| PSPNet | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 2536.944 ms | 3 - 3 MB | NPU
| PSPNet | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 2177.701 ms | 0 - 1311 MB | NPU
| PSPNet | QNN_DLC | float | Qualcomm® SA7255P | 5274.515 ms | 13 - 1329 MB | NPU
| PSPNet | QNN_DLC | float | Qualcomm® SA8295P | 1357.596 ms | 3 - 647 MB | NPU
| PSPNet | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 2177.701 ms | 0 - 1311 MB | NPU
| PSPNet | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 2340.9 ms | 0 - 1369 MB | NPU
| PSPNet | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 2104.97 ms | 1 - 1707 MB | NPU
| PSPNet | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 1856.683 ms | 1 - 934 MB | NPU
| PSPNet | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 5962.221 ms | 110 - 1507 MB | NPU
| PSPNet | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2829.444 ms | 0 - 4 MB | NPU
| PSPNet | TFLITE | float | Qualcomm® SA8775P | 2955.535 ms | 111 - 1509 MB | NPU
| PSPNet | TFLITE | float | Qualcomm® SA8650P | 2955.535 ms | 111 - 1509 MB | NPU
| PSPNet | TFLITE | float | Qualcomm® SA8255P | 2955.535 ms | 111 - 1509 MB | NPU
| PSPNet | TFLITE | float | Qualcomm® QCS8450 | 1856.683 ms | 1 - 934 MB | NPU
| PSPNet | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 2938.747 ms | 64 - 339 MB | NPU
| PSPNet | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 2172.868 ms | 1 - 1409 MB | NPU
| PSPNet | TFLITE | float | Qualcomm® SA7255P | 5962.221 ms | 110 - 1507 MB | NPU
| PSPNet | TFLITE | float | Qualcomm® SA8295P | 1414.274 ms | 128 - 837 MB | NPU
| PSPNet | TFLITE | float | Snapdragon® 8 Elite Mobile | 2172.868 ms | 1 - 1409 MB | NPU
| PSPNet | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 2329.061 ms | 0 - 1451 MB | NPU

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



## 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).