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 1.25.0	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	1.27 ms	205 - 205 MB	NPU
decoder	ONNX	float	Snapdragon® X Elite	2.106 ms	174 - 174 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	1.5 ms	0 - 272 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	2.738 ms	1 - 209 MB	NPU
decoder	ONNX	float	Qualcomm® QCS8550 (Proxy)	2.124 ms	1 - 4 MB	NPU
decoder	ONNX	float	Qualcomm® QCS8450	2.738 ms	1 - 209 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite Mobile	1.302 ms	0 - 192 MB	NPU
decoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	1.206 ms	1 - 173 MB	NPU
decoder	ONNX	float	Qualcomm® QCS9075	2.684 ms	7 - 51 MB	NPU
decoder	ONNX	float	Qualcomm® QCS8750	1.302 ms	0 - 192 MB	NPU
decoder	ONNX	float	Qualcomm® QCS7181	2.106 ms	174 - 174 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X2 Elite	1.636 ms	7 - 7 MB	NPU
decoder	QNN_DLC	float	Snapdragon® X Elite	2.229 ms	7 - 7 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	1.367 ms	0 - 274 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	2.663 ms	2 - 217 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8275	4.237 ms	5 - 105 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	2.025 ms	3 - 174 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8450	2.663 ms	2 - 217 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Mobile	1.237 ms	0 - 187 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA8295P	2.606 ms	7 - 49 MB	NPU
decoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	1.164 ms	1 - 173 MB	NPU
decoder	QNN_DLC	float	Qualcomm® SA7255P	4.237 ms	5 - 105 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS9075	2.573 ms	7 - 16 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS8750	1.237 ms	0 - 187 MB	NPU
decoder	QNN_DLC	float	Qualcomm® QCS7181	2.229 ms	7 - 7 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	1.381 ms	0 - 280 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	2.439 ms	0 - 217 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8275	4.249 ms	0 - 104 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8550 (Proxy)	1.987 ms	0 - 2 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8775P	8.85 ms	7 - 25 MB	GPU
decoder	TFLITE	float	Qualcomm® SA8650P	8.85 ms	7 - 25 MB	GPU
decoder	TFLITE	float	Qualcomm® SA8255P	8.85 ms	7 - 25 MB	GPU
decoder	TFLITE	float	Qualcomm® QCS8450	2.439 ms	0 - 217 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Mobile	1.237 ms	0 - 189 MB	NPU
decoder	TFLITE	float	Qualcomm® SA8295P	2.666 ms	0 - 42 MB	NPU
decoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	1.145 ms	0 - 175 MB	NPU
decoder	TFLITE	float	Qualcomm® SA7255P	4.249 ms	0 - 104 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS9075	2.581 ms	0 - 83 MB	NPU
decoder	TFLITE	float	Qualcomm® QCS8750	1.237 ms	0 - 189 MB	NPU
encoder	ONNX	float	Snapdragon® X2 Elite	5.484 ms	211 - 211 MB	NPU
encoder	ONNX	float	Snapdragon® X Elite	11.431 ms	180 - 180 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 3 Mobile	7.839 ms	16 - 205 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Gen 1 Mobile	20.027 ms	16 - 322 MB	NPU
encoder	ONNX	float	Qualcomm® QCS8550 (Proxy)	10.915 ms	0 - 59 MB	NPU
encoder	ONNX	float	Qualcomm® QCS8450	20.027 ms	16 - 322 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Mobile	5.294 ms	16 - 141 MB	NPU
encoder	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	4.13 ms	2 - 124 MB	NPU
encoder	ONNX	float	Qualcomm® QCS9075	14.176 ms	15 - 60 MB	NPU
encoder	ONNX	float	Qualcomm® QCS8750	5.294 ms	16 - 141 MB	NPU
encoder	ONNX	float	Qualcomm® QCS7181	11.431 ms	180 - 180 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X2 Elite	5.178 ms	2 - 2 MB	NPU
encoder	QNN_DLC	float	Snapdragon® X Elite	12.181 ms	2 - 2 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	8.016 ms	0 - 190 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	19.436 ms	0 - 306 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8275	37.03 ms	2 - 107 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	11.295 ms	2 - 170 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8450	19.436 ms	0 - 306 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Mobile	5.448 ms	2 - 113 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA8295P	19.688 ms	2 - 233 MB	NPU
encoder	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	4.328 ms	2 - 116 MB	NPU
encoder	QNN_DLC	float	Qualcomm® SA7255P	37.03 ms	2 - 107 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS9075	15.108 ms	2 - 12 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS8750	5.448 ms	2 - 113 MB	NPU
encoder	QNN_DLC	float	Qualcomm® QCS7181	12.181 ms	2 - 2 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	7.705 ms	3 - 195 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	19.541 ms	7 - 311 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS8275	36.805 ms	7 - 119 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS8550 (Proxy)	10.741 ms	7 - 9 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8775P	122.908 ms	6 - 34 MB	GPU
encoder	TFLITE	float	Qualcomm® SA8650P	122.908 ms	6 - 34 MB	GPU
encoder	TFLITE	float	Qualcomm® SA8255P	122.908 ms	6 - 34 MB	GPU
encoder	TFLITE	float	Qualcomm® QCS8450	19.541 ms	7 - 311 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Elite Mobile	5.202 ms	6 - 118 MB	NPU
encoder	TFLITE	float	Qualcomm® SA8295P	19.12 ms	7 - 236 MB	NPU
encoder	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	4.102 ms	0 - 105 MB	NPU
encoder	TFLITE	float	Qualcomm® SA7255P	36.805 ms	7 - 119 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS9075	13.921 ms	5 - 65 MB	NPU
encoder	TFLITE	float	Qualcomm® QCS8750	5.202 ms	6 - 118 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