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 2.42, ONNX Runtime 1.24.3	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.19.1	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® 8 Elite Gen 5 Mobile	1.142 ms	0 - 224 MB	NPU
decoder	ONNX	float	Snapdragon® X2 Elite	1.153 ms	68 - 68 MB	NPU
decoder	ONNX	float	Snapdragon® X Elite	2.348 ms	67 - 67 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.456 ms	0 - 258 MB	NPU
decoder	ONNX	float	Qualcomm® QCS8550 (Proxy)	2.138 ms	1 - 4 MB	NPU
decoder	ONNX	float	Qualcomm® QCS9075	2.714 ms	7 - 16 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	1.248 ms	0 - 252 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	1.117 ms	1 - 212 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X2 Elite	1.6 ms	7 - 7 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X Elite	2.141 ms	7 - 7 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1.403 ms	0 - 232 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	4.216 ms	7 - 143 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	1.952 ms	1 - 3 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8775P	2.833 ms	7 - 144 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS9075	2.549 ms	7 - 16 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	2.869 ms	1 - 215 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA7255P	4.216 ms	7 - 143 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8295P	2.934 ms	7 - 130 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	1.236 ms	0 - 226 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1.097 ms	0 - 224 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1.36 ms	0 - 247 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8275 (Proxy)	4.235 ms	0 - 155 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8550 (Proxy)	1.939 ms	0 - 3 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8775P	2.799 ms	0 - 151 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS9075	2.514 ms	0 - 83 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8450 (Proxy)	2.575 ms	0 - 225 MB	NPU
decoder	TFLITE	float	Qualcomm® SA7255P	4.235 ms	0 - 155 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8295P	3.039 ms	0 - 134 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	1.164 ms	0 - 231 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	4.389 ms	1 - 178 MB	NPU
encoder	ONNX	float	Snapdragon® X2 Elite	4.863 ms	48 - 48 MB	NPU
encoder	ONNX	float	Snapdragon® X Elite	12.577 ms	47 - 47 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	8.29 ms	16 - 269 MB	NPU
encoder	ONNX	float	Qualcomm® QCS8550 (Proxy)	11.99 ms	0 - 60 MB	NPU
encoder	ONNX	float	Qualcomm® QCS9075	15.758 ms	15 - 19 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	5.752 ms	16 - 190 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	4.662 ms	2 - 156 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X2 Elite	5.659 ms	2 - 2 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X Elite	12.47 ms	2 - 2 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	8.248 ms	0 - 231 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	38.122 ms	2 - 149 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	11.819 ms	2 - 3 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8775P	14.225 ms	2 - 149 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS9075	16.741 ms	2 - 12 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	21.927 ms	0 - 345 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA7255P	38.122 ms	2 - 149 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8295P	20.266 ms	2 - 273 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	5.765 ms	2 - 155 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	4.067 ms	0 - 145 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	7.719 ms	6 - 227 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS8275 (Proxy)	36.775 ms	7 - 145 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS8550 (Proxy)	10.703 ms	7 - 9 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8775P	13.375 ms	7 - 146 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS9075	13.963 ms	6 - 65 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS8450 (Proxy)	19.604 ms	7 - 340 MB	NPU
encoder	TFLITE	float	Qualcomm® SA7255P	36.775 ms	7 - 145 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8295P	19.123 ms	7 - 269 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	5.219 ms	5 - 153 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