TrOCR: Optimized for Qualcomm Devices
End-to-end text recognition approach with pre-trained image transformer and text transformer models for both image understanding and wordpiece-level text generation.
This is based on the implementation of TrOCR 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 2.51, ONNX Runtime 1.30.0 | Download |
| QNN_DLC | float | Universal | QAIRT 2.51 | Download |
| TFLITE | float | Universal | QAIRT 2.51 | Download |
For more device-specific assets and performance metrics, visit TrOCR 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 TrOCR on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.image_to_text
Model Stats:
- Input resolution: 320x320
- Model checkpoint: trocr-small-handwritten
- Model size (decoder) (float): 146 MB
- Model size (encoder) (float): 87.8 MB
- Number of parameters (decoder): 38.3M
- Number of parameters (encoder): 23.0M
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| decoder | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 1.29 ms | 2 - 181 MB | NPU |
| decoder | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.252 ms | 0 - 193 MB | NPU |
| decoder | ONNX | float | Snapdragon® X2 Elite | 1.266 ms | 7 - 7 MB | NPU |
| decoder | ONNX | float | Snapdragon® X Elite | 2.008 ms | 67 - 67 MB | NPU |
| decoder | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 1.479 ms | 0 - 283 MB | NPU |
| decoder | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 2.356 ms | 1 - 224 MB | NPU |
| decoder | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 2.545 ms | 7 - 18 MB | NPU |
| decoder | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2.084 ms | 0 - 75 MB | NPU |
| decoder | ONNX | float | Qualcomm® QCS8450 | 2.356 ms | 1 - 224 MB | NPU |
| decoder | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 2.57 ms | 7 - 17 MB | NPU |
| decoder | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 2.008 ms | 67 - 67 MB | NPU |
| decoder | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 1.252 ms | 0 - 193 MB | NPU |
| decoder | ONNX | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 1.244 ms | 1 - 195 MB | NPU |
| decoder | ONNX | float | Snapdragon® 8 Elite Gen 6 Mobile | 1.231 ms | 0 - 109 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 1.266 ms | 1 - 177 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.235 ms | 0 - 189 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® X2 Elite | 1.687 ms | 7 - 7 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® X Elite | 2.241 ms | 7 - 7 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 1.466 ms | 0 - 280 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 2.268 ms | 2 - 223 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 2.431 ms | 7 - 17 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.982 ms | 2 - 4 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® SA8775P | 2.801 ms | 1 - 105 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® SA8650P | 2.801 ms | 1 - 105 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® SA8255P | 2.801 ms | 1 - 105 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® QCS8450 | 2.268 ms | 2 - 223 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 2.371 ms | 7 - 16 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 2.241 ms | 7 - 7 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 1.235 ms | 0 - 189 MB | NPU |
| decoder | QNN_DLC | float | Qualcomm® SA8295P | 2.562 ms | 6 - 53 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 1.249 ms | 1 - 191 MB | NPU |
| decoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 6 Mobile | 1.279 ms | 1 - 106 MB | NPU |
| decoder | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 1.269 ms | 0 - 173 MB | NPU |
| decoder | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.289 ms | 0 - 192 MB | NPU |
| decoder | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.437 ms | 0 - 282 MB | NPU |
| decoder | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 2.27 ms | 0 - 224 MB | NPU |
| decoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 2.411 ms | 0 - 84 MB | NPU |
| decoder | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2.036 ms | 0 - 2 MB | NPU |
| decoder | TFLITE | float | Qualcomm® SA8775P | 2.772 ms | 0 - 105 MB | NPU |
| decoder | TFLITE | float | Qualcomm® SA8650P | 2.772 ms | 0 - 105 MB | NPU |
| decoder | TFLITE | float | Qualcomm® SA8255P | 2.772 ms | 0 - 105 MB | NPU |
| decoder | TFLITE | float | Qualcomm® QCS8450 | 2.27 ms | 0 - 224 MB | NPU |
| decoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 2.341 ms | 0 - 84 MB | NPU |
| decoder | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 1.289 ms | 0 - 192 MB | NPU |
| decoder | TFLITE | float | Qualcomm® SA7255P | 6.765 ms | 7 - 20 MB | CPU |
| decoder | TFLITE | float | Qualcomm® SA8295P | 2.524 ms | 0 - 49 MB | NPU |
| decoder | TFLITE | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 1.161 ms | 0 - 193 MB | NPU |
| decoder | TFLITE | float | Snapdragon® 8 Elite Gen 6 Mobile | 1.307 ms | 0 - 107 MB | NPU |
| encoder | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 4.556 ms | 14 - 143 MB | NPU |
| encoder | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 5.578 ms | 15 - 141 MB | NPU |
| encoder | ONNX | float | Snapdragon® X2 Elite | 4.949 ms | 16 - 16 MB | NPU |
| encoder | ONNX | float | Snapdragon® X Elite | 11.872 ms | 48 - 48 MB | NPU |
| encoder | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 8.089 ms | 0 - 206 MB | NPU |
| encoder | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 18.989 ms | 0 - 317 MB | NPU |
| encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 14.232 ms | 15 - 20 MB | NPU |
| encoder | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 11.311 ms | 0 - 56 MB | NPU |
| encoder | ONNX | float | Qualcomm® QCS8450 | 18.989 ms | 0 - 317 MB | NPU |
| encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 13.706 ms | 15 - 20 MB | NPU |
| encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 11.872 ms | 48 - 48 MB | NPU |
| encoder | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 5.578 ms | 15 - 141 MB | NPU |
| encoder | ONNX | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 4.016 ms | 1 - 129 MB | NPU |
| encoder | ONNX | float | Snapdragon® 8 Elite Gen 6 Mobile | 5.302 ms | 15 - 140 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 4.709 ms | 2 - 119 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 5.711 ms | 2 - 116 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® X2 Elite | 5.522 ms | 2 - 2 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® X Elite | 12.606 ms | 2 - 2 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 7.686 ms | 2 - 194 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 19.621 ms | 2 - 309 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 14.502 ms | 2 - 13 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 11.837 ms | 2 - 4 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® SA8775P | 14.365 ms | 2 - 112 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® SA8650P | 14.365 ms | 2 - 112 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® SA8255P | 14.365 ms | 2 - 112 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® QCS8450 | 19.621 ms | 2 - 309 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 14.13 ms | 4 - 15 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 12.606 ms | 2 - 2 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 5.711 ms | 2 - 116 MB | NPU |
| encoder | QNN_DLC | float | Qualcomm® SA8295P | 20.083 ms | 2 - 225 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 4.121 ms | 2 - 117 MB | NPU |
| encoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 6 Mobile | 5.519 ms | 2 - 112 MB | NPU |
| encoder | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 4.067 ms | 0 - 105 MB | NPU |
| encoder | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.951 ms | 6 - 115 MB | NPU |
| encoder | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 7.382 ms | 5 - 196 MB | NPU |
| encoder | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 16.984 ms | 7 - 309 MB | NPU |
| encoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 13.069 ms | 7 - 68 MB | NPU |
| encoder | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 10.141 ms | 7 - 10 MB | NPU |
| encoder | TFLITE | float | Qualcomm® SA8775P | 12.824 ms | 7 - 111 MB | NPU |
| encoder | TFLITE | float | Qualcomm® SA8650P | 12.824 ms | 7 - 111 MB | NPU |
| encoder | TFLITE | float | Qualcomm® SA8255P | 12.824 ms | 7 - 111 MB | NPU |
| encoder | TFLITE | float | Qualcomm® QCS8450 | 16.984 ms | 7 - 309 MB | NPU |
| encoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 12.873 ms | 6 - 66 MB | NPU |
| encoder | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 4.951 ms | 6 - 115 MB | NPU |
| encoder | TFLITE | float | Qualcomm® SA7255P | 353.772 ms | 8 - 21 MB | CPU |
| encoder | TFLITE | float | Qualcomm® SA8295P | 18.208 ms | 7 - 222 MB | NPU |
| encoder | TFLITE | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 3.608 ms | 0 - 109 MB | NPU |
| encoder | TFLITE | float | Snapdragon® 8 Elite Gen 6 Mobile | 4.822 ms | 6 - 114 MB | NPU |
License
- The license for the original implementation of TrOCR can be found here.
References
- TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models
- 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.
