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 2.50, ONNX Runtime 1.27.1	Download
QNN_DLC	float	Universal	QAIRT 2.50	Download
TFLITE	float	Universal	QAIRT 2.50	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	1.304 ms	2 - 184 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	1.282 ms	0 - 196 MB	NPU
decoder	ONNX	float	Snapdragon® X2 Elite	1.267 ms	7 - 7 MB	NPU
decoder	ONNX	float	Snapdragon® X Elite	1.931 ms	68 - 68 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.512 ms	0 - 285 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	2.313 ms	1 - 224 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	2.811 ms	7 - 18 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	2.106 ms	1 - 4 MB	NPU
decoder	ONNX	float	Qualcomm® QCS8450	2.313 ms	1 - 224 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	2.759 ms	7 - 17 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	1.931 ms	68 - 68 MB	NPU
decoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	1.282 ms	0 - 196 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	1.278 ms	2 - 178 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	1.257 ms	0 - 189 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X2 Elite	1.701 ms	7 - 7 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X Elite	2.156 ms	7 - 7 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1.449 ms	0 - 278 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	2.277 ms	2 - 223 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	2.661 ms	7 - 17 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	2.015 ms	1 - 3 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8650P	2.892 ms	5 - 106 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8255P	2.892 ms	5 - 106 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8450	2.277 ms	2 - 223 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	2.602 ms	7 - 16 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	2.156 ms	7 - 7 MB	NPU
decoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	1.257 ms	0 - 189 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA7255P	4.13 ms	7 - 110 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8295P	2.595 ms	0 - 45 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	1.257 ms	0 - 179 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	1.239 ms	0 - 192 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1.42 ms	18 - 301 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	2.271 ms	0 - 220 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	2.665 ms	0 - 84 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	1.98 ms	0 - 3 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8650P	2.84 ms	0 - 104 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8255P	2.84 ms	0 - 104 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8450	2.271 ms	0 - 220 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	2.582 ms	0 - 83 MB	NPU
decoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	1.239 ms	0 - 192 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8295P	2.558 ms	0 - 43 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	4.111 ms	2 - 126 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	5.119 ms	16 - 141 MB	NPU
encoder	ONNX	float	Snapdragon® X2 Elite	4.481 ms	16 - 16 MB	NPU
encoder	ONNX	float	Snapdragon® X Elite	10.799 ms	48 - 48 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	7.491 ms	0 - 203 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	17.43 ms	0 - 313 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	13.562 ms	15 - 20 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	10.388 ms	0 - 57 MB	NPU
encoder	ONNX	float	Qualcomm® QCS8450	17.43 ms	0 - 313 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	12.92 ms	15 - 20 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	10.799 ms	48 - 48 MB	NPU
encoder	ONNX	float	Qualcomm® Dragonwing™ Q-8750	5.119 ms	16 - 141 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	4.269 ms	2 - 116 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	5.263 ms	1 - 112 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X2 Elite	5.041 ms	2 - 2 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X Elite	11.705 ms	2 - 2 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	7.697 ms	2 - 194 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	17.931 ms	0 - 303 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	13.597 ms	2 - 13 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	10.93 ms	2 - 157 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8650P	14.024 ms	2 - 108 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8255P	14.024 ms	2 - 108 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8450	17.931 ms	0 - 303 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	13.328 ms	2 - 12 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	11.705 ms	2 - 2 MB	NPU
encoder	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	5.263 ms	1 - 112 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA7255P	36.097 ms	2 - 108 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8295P	18.992 ms	2 - 221 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	5.007 ms	5 - 115 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	7.365 ms	5 - 201 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	17.221 ms	5 - 303 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	13.23 ms	7 - 67 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	10.284 ms	7 - 9 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8650P	13.479 ms	7 - 111 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8255P	13.479 ms	7 - 111 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS8450	17.221 ms	5 - 303 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	12.79 ms	6 - 66 MB	NPU
encoder	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	5.007 ms	5 - 115 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8295P	18.385 ms	7 - 222 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