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 1.25.0 | Download |
| ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.25.0 | 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 | 686.741 ms | 38 - 2698 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Snapdragon® X2 Elite | 639.291 ms | 188 - 188 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 889.706 ms | 47 - 3964 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Qualcomm® QCS8550 (Proxy) | 1221.859 ms | 0 - 182 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 781.669 ms | 12 - 2549 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Qualcomm® QCS9075 | 1389.741 ms | 24 - 93 MB | NPU |
| HRNet-W48-OCR | ONNX | float | Qualcomm® QCS8750 | 781.669 ms | 12 - 2549 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 5037.067 ms | 15 - 2141 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® X2 Elite | 5959.77 ms | 153 - 153 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® X Elite | 8852.457 ms | 169 - 169 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 6839.983 ms | 14 - 3331 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Qualcomm® QCS6490 | 56100.685 ms | 2501 - 2505 MB | CPU |
| HRNet-W48-OCR | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 9039.201 ms | 1 - 88 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Qualcomm® QCM6690 | 27535.857 ms | 2479 - 2499 MB | CPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 28009.483 ms | 2443 - 2459 MB | CPU |
| HRNet-W48-OCR | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 9996.686 ms | 3 - 1937 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Qualcomm® QCS9075 | 8837.461 ms | 10 - 67 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Qualcomm® QCS7790 | 28009.483 ms | 2443 - 2459 MB | CPU |
| HRNet-W48-OCR | ONNX | w8a16 | Qualcomm® QCS8750 | 9996.686 ms | 3 - 1937 MB | NPU |
| HRNet-W48-OCR | ONNX | w8a16 | Qualcomm® QCS7181 | 8852.457 ms | 169 - 169 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 5082.428 ms | 12 - 2073 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 6823.781 ms | 12 - 12 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® X Elite | 46721.81 ms | 12 - 12 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 11427.611 ms | 12 - 2743 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® QCS8275 | 50342.001 ms | 12 - 1685 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 50456.2 ms | 12 - 16 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® SA8775P | 53284.416 ms | 12 - 1683 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® SA8650P | 53284.416 ms | 12 - 1683 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® SA8255P | 53284.416 ms | 12 - 1683 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 218505.325 ms | 29 - 2967 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® SA7255P | 50342.001 ms | 12 - 1685 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 44244.936 ms | 15 - 2055 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 10350.323 ms | 12 - 1872 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 71730.154 ms | 12 - 30 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® QCS7790 | 44244.936 ms | 15 - 2055 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® QCS8750 | 10350.323 ms | 12 - 1872 MB | NPU |
| HRNet-W48-OCR | QNN_DLC | w8a16 | Qualcomm® QCS7181 | 46721.81 ms | 12 - 12 MB | NPU |
| HRNet-W48-OCR | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 14976.461 ms | 5 - 304 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® QCS8275 | 28138.973 ms | 6 - 287 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 18998.575 ms | 15 - 21 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® SA8775P | 19948.822 ms | 6 - 287 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® SA8650P | 19948.822 ms | 6 - 287 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® SA8255P | 19948.822 ms | 6 - 287 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 26208.445 ms | 20 - 361 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 10698.763 ms | 12 - 306 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® QCS9075 | 18837.509 ms | 8 - 2686 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® SA7255P | 28138.973 ms | 6 - 287 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® SA8295P | 16626.77 ms | 8 - 335 MB | CPU |
| HRNet-W48-OCR | TFLITE | float | Qualcomm® QCS8750 | 10698.763 ms | 12 - 306 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.