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 2.45, ONNX Runtime 1.27.1	Download
QNN_DLC	float	Universal	QAIRT 2.45	Download
TFLITE	float	Universal	QAIRT 2.45	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:

Model checkpoint: trocr-small-stage1
Input resolution: 320x320
Number of parameters (decoder): 38.3M
Model size (decoder) (float): 146 MB
Number of parameters (encoder): 23.0M
Model size (encoder) (float): 87.8 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
decoder	ONNX	float	Snapdragon® X2 Elite	1.258 ms	7 - 7 MB	NPU
decoder	ONNX	float	Snapdragon® X Elite	2.061 ms	68 - 68 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.452 ms	0 - 283 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	2.746 ms	1 - 218 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	2.12 ms	3 - 238 MB	NPU
decoder	ONNX	float	Qualcomm® QCS8450	2.746 ms	1 - 218 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	2.759 ms	7 - 16 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	2.061 ms	68 - 68 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	1.288 ms	0 - 191 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite Mobile	1.288 ms	0 - 191 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1.172 ms	1 - 177 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X2 Elite	1.57 ms	7 - 7 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X Elite	2.228 ms	7 - 7 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1.368 ms	0 - 275 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	2.698 ms	2 - 217 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8275	4.177 ms	5 - 105 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	2.003 ms	4 - 6 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8775P	2.853 ms	5 - 106 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8650P	2.853 ms	5 - 106 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8255P	2.853 ms	5 - 106 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8450	2.698 ms	2 - 217 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	2.629 ms	7 - 15 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	2.228 ms	7 - 7 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	1.223 ms	0 - 186 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA7255P	4.177 ms	5 - 105 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8295P	2.602 ms	7 - 49 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Mobile	1.223 ms	0 - 186 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	1.151 ms	1 - 175 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1.395 ms	0 - 280 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	2.442 ms	0 - 218 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8275	4.237 ms	0 - 103 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.96 ms	0 - 3 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8775P	2.865 ms	0 - 103 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8650P	2.865 ms	0 - 103 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8255P	2.865 ms	0 - 103 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8450	2.442 ms	0 - 218 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	2.585 ms	0 - 83 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	1.243 ms	0 - 192 MB	NPU
decoder	TFLITE	float	Qualcomm® SA7255P	4.237 ms	0 - 103 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8295P	2.647 ms	0 - 42 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Mobile	1.243 ms	0 - 192 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1.127 ms	0 - 172 MB	NPU
encoder	ONNX	float	Snapdragon® X2 Elite	4.546 ms	16 - 16 MB	NPU
encoder	ONNX	float	Snapdragon® X Elite	11.355 ms	48 - 48 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	7.819 ms	0 - 198 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	19.911 ms	0 - 315 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	10.914 ms	0 - 57 MB	NPU
encoder	ONNX	float	Qualcomm® QCS8450	19.911 ms	0 - 315 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	13.326 ms	15 - 19 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	11.355 ms	48 - 48 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	5.303 ms	16 - 140 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Mobile	5.303 ms	16 - 140 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	4.146 ms	0 - 124 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X2 Elite	5.17 ms	2 - 2 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X Elite	12.208 ms	2 - 2 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	8.013 ms	0 - 188 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	19.533 ms	0 - 307 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8275	37.065 ms	2 - 107 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	11.299 ms	2 - 176 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8775P	14.043 ms	2 - 109 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8650P	14.043 ms	2 - 109 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8255P	14.043 ms	2 - 109 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8450	19.533 ms	0 - 307 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	14.564 ms	2 - 12 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	12.208 ms	2 - 2 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	5.444 ms	2 - 114 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA7255P	37.065 ms	2 - 107 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8295P	19.696 ms	2 - 235 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Mobile	5.444 ms	2 - 114 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	4.345 ms	2 - 115 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	7.708 ms	5 - 194 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	19.583 ms	7 - 312 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8275	36.81 ms	7 - 119 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	10.695 ms	7 - 9 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8775P	13.5 ms	7 - 110 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8650P	13.5 ms	7 - 110 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8255P	13.5 ms	7 - 110 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS8450	19.583 ms	7 - 312 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	13.517 ms	6 - 66 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	5.177 ms	7 - 119 MB	NPU
encoder	TFLITE	float	Qualcomm® SA7255P	36.81 ms	7 - 119 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8295P	19.131 ms	7 - 237 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Elite Mobile	5.177 ms	7 - 119 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	4.115 ms	0 - 106 MB	NPU

License

The license for the original implementation of TrOCR can be found here.

References

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.

Downloads last month: -; Downloads are not tracked for this model. How to track

Space using qualcomm/TrOCR 1

Paper for qualcomm/TrOCR

TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models

Paper • 2109.10282 • Published Sep 21, 2021 • 13