--- 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/main/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 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pspnet/releases/v0.49.1/pspnet-onnx-float.zip) | ONNX | w8a8 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pspnet/releases/v0.49.1/pspnet-onnx-w8a8.zip) | QNN_DLC | float | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pspnet/releases/v0.49.1/pspnet-qnn_dlc-float.zip) | QNN_DLC | w8a8 | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pspnet/releases/v0.49.1/pspnet-qnn_dlc-w8a8.zip) | TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pspnet/releases/v0.49.1/pspnet-tflite-float.zip) | TFLITE | w8a8 | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/pspnet/releases/v0.49.1/pspnet-tflite-w8a8.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/main/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/main/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® 8 Elite Gen 5 Mobile | 323.654 ms | 233 - 1001 MB | NPU | PSPNet | ONNX | float | Snapdragon® X2 Elite | 427.869 ms | 266 - 266 MB | NPU | PSPNet | ONNX | float | Snapdragon® X Elite | 695.005 ms | 529 - 529 MB | NPU | PSPNet | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 506.756 ms | 136 - 1053 MB | NPU | PSPNet | ONNX | float | Qualcomm® QCS8550 (Proxy) | 602.812 ms | 195 - 199 MB | NPU | PSPNet | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 333.199 ms | 246 - 1003 MB | NPU | PSPNet | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 16.884 ms | 72 - 340 MB | NPU | PSPNet | ONNX | w8a8 | Snapdragon® X2 Elite | 15.251 ms | 131 - 131 MB | NPU | PSPNet | ONNX | w8a8 | Snapdragon® X Elite | 28.661 ms | 133 - 133 MB | NPU | PSPNet | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 22.334 ms | 72 - 494 MB | NPU | PSPNet | ONNX | w8a8 | Qualcomm® QCS6490 | 3318.926 ms | 196 - 276 MB | CPU | PSPNet | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 29.465 ms | 64 - 576 MB | NPU | PSPNet | ONNX | w8a8 | Qualcomm® QCS9075 | 34.444 ms | 70 - 74 MB | NPU | PSPNet | ONNX | w8a8 | Qualcomm® QCM6690 | 3277.178 ms | 95 - 110 MB | CPU | PSPNet | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 19.038 ms | 72 - 343 MB | NPU | PSPNet | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 2589.577 ms | 50 - 62 MB | CPU | PSPNet | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 269.084 ms | 3 - 741 MB | NPU | PSPNet | QNN_DLC | float | Snapdragon® X2 Elite | 229.121 ms | 3 - 3 MB | NPU | PSPNet | QNN_DLC | float | Snapdragon® X Elite | 545.535 ms | 3 - 3 MB | NPU | PSPNet | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 623.685 ms | 0 - 874 MB | NPU | PSPNet | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 1863.358 ms | 2 - 722 MB | NPU | PSPNet | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 608.77 ms | 3 - 5 MB | NPU | PSPNet | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 1689.881 ms | 0 - 431 MB | NPU | PSPNet | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 301.937 ms | 0 - 718 MB | NPU | PSPNet | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 16.214 ms | 1 - 239 MB | NPU | PSPNet | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 14.344 ms | 1 - 1 MB | NPU | PSPNet | QNN_DLC | w8a8 | Snapdragon® X Elite | 26.117 ms | 1 - 1 MB | NPU | PSPNet | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 20.679 ms | 1 - 365 MB | NPU | PSPNet | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 63.897 ms | 1 - 263 MB | NPU | PSPNet | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 27.013 ms | 0 - 9 MB | NPU | PSPNet | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 31.865 ms | 0 - 34 MB | NPU | PSPNet | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 36.677 ms | 1 - 358 MB | NPU | PSPNet | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 18.068 ms | 1 - 234 MB | NPU | PSPNet | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 91.895 ms | 1 - 458 MB | NPU | PSPNet | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 326.397 ms | 128 - 965 MB | NPU | PSPNet | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 528.08 ms | 0 - 1049 MB | NPU | PSPNet | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 1688.514 ms | 124 - 958 MB | NPU | PSPNet | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 619.017 ms | 128 - 131 MB | NPU | PSPNet | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 1563.118 ms | 125 - 698 MB | NPU | PSPNet | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 974.903 ms | 128 - 884 MB | NPU | PSPNet | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 17.665 ms | 32 - 284 MB | NPU | PSPNet | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 26.587 ms | 30 - 427 MB | NPU | PSPNet | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 74.606 ms | 32 - 304 MB | NPU | PSPNet | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 34.862 ms | 32 - 35 MB | NPU | PSPNet | TFLITE | w8a8 | Qualcomm® QCS9075 | 38.16 ms | 32 - 131 MB | NPU | PSPNet | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 55.328 ms | 32 - 417 MB | NPU | PSPNet | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 22.633 ms | 32 - 286 MB | NPU | PSPNet | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 356.021 ms | 384 - 904 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).