Segmentation Transformer: Object-Contextual Representations for Semantic Segmentation
Paper • 1909.11065 • Published
HRNet-W48-OCR: Optimized for Qualcomm Devices
HRNet-W48-OCR is a machine learning model that can segment images from the Cityscape dataset. It has lightweight and hardware-efficient operations and thus delivers significant speedup on diverse hardware platforms
This is based on the implementation of HRNet-W48-OCR found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.
Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up 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 1.24.3 | Download |
| ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| QNN_DLC | w8a16 | Universal | QAIRT 2.45 | Download |
| TFLITE | float | Universal | Download |
For more device-specific assets and performance metrics, visit HRNet-W48-OCR on Qualcomm® AI Hub.
Option 2: Export with Custom Configurations
Use the Qualcomm® AI Hub Models 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 HRNet-W48-OCR on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.semantic_segmentation
Model Stats:
- Model checkpoint: hrnet_ocr_cs_8162_torch11.pth
- Input resolution: 2048x1024
- Number of output classes: 19
- Number of parameters: 70.3M
- Model size (float): 268 MB
- Model size (w8a16): 70.3 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| HRNet-W48-OCR | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 673.675 ms | 37 - 2715 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Snapdragon® 8 Elite Mobile | 785.197 ms | 12 - 2542 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Snapdragon® X2 Elite | 665.888 ms | 148 - 148 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Snapdragon® X Elite | 1093.749 ms | 146 - 146 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Snapdragon® X Elite | 1093.749 ms | 146 - 146 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 914.325 ms | 42 - 3961 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Qualcomm® QCS8550 (Proxy) | 1235.984 ms | 0 - 168 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 785.197 ms | 12 - 2542 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Qualcomm® QCS9075 | 1388.628 ms | 24 - 51 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 4975.488 ms | 29 - 2965 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 10444.216 ms | 16 - 2522 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® X2 Elite | 6061.017 ms | 90 - 90 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® X Elite | 8417.311 ms | 84 - 84 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® X Elite | 8417.311 ms | 84 - 84 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 6897.347 ms | 30 - 4077 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Qualcomm® QCS6490 | 55892.053 ms | 2501 - 2505 MB | CPU |
| HRNet-W48-OCR | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 8776.629 ms | 26 - 34 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 10444.216 ms | 16 - 2522 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 27997.122 ms | 2444 - 2459 MB | CPU |
| HRNet-W48-OCR | ONNX | w8a16 | Qualcomm® QCS9075 | 8976.655 ms | 21 - 35 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Qualcomm® QCM6690 | 27579.967 ms | 2479 - 2498 MB | CPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 27997.122 ms | 2444 - 2459 MB | CPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 4989.952 ms | 14 - 2075 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 10359.81 ms | 12 - 1871 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 6948.17 ms | 12 - 12 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® X Elite | 42385.074 ms | 12 - 12 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® X Elite | 42385.074 ms | 12 - 12 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 11408.25 ms | 12 - 2745 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 50479.47 ms | 13 - 15 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® SA8775P | 53377.121 ms | 12 - 1684 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® SA8775P | 53377.121 ms | 12 - 1684 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® SA8775P | 53377.121 ms | 12 - 1684 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® SA7255P | 50878.724 ms | 12 - 1686 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 10359.81 ms | 12 - 1871 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 44305.56 ms | 12 - 2050 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 71825.471 ms | 12 - 30 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 44305.56 ms | 12 - 2050 MB | NPU |
| HRNet-W48-OCR | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 9396.833 ms | 5 - 320 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Snapdragon® 8 Elite Mobile | 14418.188 ms | 8 - 296 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 14999.933 ms | 2 - 300 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 17713.601 ms | 12 - 18 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® SA8775P | 19770.945 ms | 4 - 285 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® SA8775P | 19770.945 ms | 4 - 285 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® SA8775P | 19770.945 ms | 4 - 285 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® SA8295P | 15984.71 ms | 11 - 338 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 14418.188 ms | 8 - 296 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® QCS9075 | 18837.819 ms | 2 - 2680 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 25992.885 ms | 10 - 354 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® SA7255P | 26715.001 ms | 5 - 53 MB | CPU |
License
- The license for the original implementation of HRNet-W48-OCR can be found here.
References
- Segmentation Transformer: Object-Contextual Representations for Semantic Segmentation
- Source Model Implementation
Community
- Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
- For questions or feedback please reach out to us.