TrOCR: Optimized for Mobile Deployment

Transformer based model for state-of-the-art optical character recognition (OCR) on both printed and handwritten text

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 model is an implementation of TrOCR found here.

This repository provides scripts to run TrOCR on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

Model Type: Model_use_case.image_to_text
Model Stats:
- Model checkpoint: trocr-small-stage1
- Input resolution: 320x320
- Number of parameters (TrOCRDecoder): 38.3M
- Model size (TrOCRDecoder) (float): 146 MB
- Number of parameters (TrOCREncoder): 23.0M
- Model size (TrOCREncoder) (float): 87.8 MB

Model	Precision	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit	Target Model
TrOCRDecoder	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	TFLITE	4.155 ms	0 - 222 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN_DLC	4.189 ms	0 - 213 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	TFLITE	2.814 ms	0 - 296 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN_DLC	2.832 ms	1 - 292 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	TFLITE	2.02 ms	0 - 3 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_DLC	1.927 ms	2 - 4 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	ONNX	2.499 ms	1 - 4 MB	NPU	TrOCR.onnx.zip
TrOCRDecoder	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	TFLITE	2.881 ms	0 - 219 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_DLC	2.796 ms	7 - 217 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	SA7255P ADP	Qualcomm® SA7255P	TFLITE	4.155 ms	0 - 222 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	SA7255P ADP	Qualcomm® SA7255P	QNN_DLC	4.189 ms	0 - 213 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	TFLITE	2.035 ms	0 - 3 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_DLC	1.947 ms	2 - 5 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	SA8295P ADP	Qualcomm® SA8295P	TFLITE	2.817 ms	0 - 207 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	SA8295P ADP	Qualcomm® SA8295P	QNN_DLC	2.817 ms	6 - 207 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	TFLITE	2.013 ms	0 - 3 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_DLC	1.932 ms	1 - 3 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	SA8775P ADP	Qualcomm® SA8775P	TFLITE	2.881 ms	0 - 219 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	SA8775P ADP	Qualcomm® SA8775P	QNN_DLC	2.796 ms	7 - 217 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	TFLITE	1.407 ms	0 - 319 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_DLC	1.415 ms	0 - 311 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	1.765 ms	0 - 290 MB	NPU	TrOCR.onnx.zip
TrOCRDecoder	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	TFLITE	1.157 ms	0 - 303 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_DLC	1.156 ms	0 - 294 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	ONNX	1.512 ms	0 - 268 MB	NPU	TrOCR.onnx.zip
TrOCRDecoder	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	TFLITE	1.099 ms	0 - 282 MB	NPU	TrOCR.tflite
TrOCRDecoder	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_DLC	1.11 ms	2 - 279 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	ONNX	1.378 ms	2 - 259 MB	NPU	TrOCR.onnx.zip
TrOCRDecoder	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_DLC	2.174 ms	7 - 7 MB	NPU	TrOCR.dlc
TrOCRDecoder	float	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	2.265 ms	68 - 68 MB	NPU	TrOCR.onnx.zip
TrOCREncoder	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	TFLITE	38.054 ms	7 - 252 MB	NPU	TrOCR.tflite
TrOCREncoder	float	QCS8275 (Proxy)	Qualcomm® QCS8275 (Proxy)	QNN_DLC	45.721 ms	2 - 260 MB	NPU	TrOCR.dlc
TrOCREncoder	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	TFLITE	21.572 ms	7 - 349 MB	NPU	TrOCR.tflite
TrOCREncoder	float	QCS8450 (Proxy)	Qualcomm® QCS8450 (Proxy)	QNN_DLC	28.194 ms	2 - 356 MB	NPU	TrOCR.dlc
TrOCREncoder	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	TFLITE	12.106 ms	7 - 9 MB	NPU	TrOCR.tflite
TrOCREncoder	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_DLC	16.659 ms	2 - 4 MB	NPU	TrOCR.dlc
TrOCREncoder	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	ONNX	17.607 ms	0 - 58 MB	NPU	TrOCR.onnx.zip
TrOCREncoder	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	TFLITE	14.633 ms	7 - 252 MB	NPU	TrOCR.tflite
TrOCREncoder	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_DLC	19.802 ms	2 - 263 MB	NPU	TrOCR.dlc
TrOCREncoder	float	SA7255P ADP	Qualcomm® SA7255P	TFLITE	38.054 ms	7 - 252 MB	NPU	TrOCR.tflite
TrOCREncoder	float	SA7255P ADP	Qualcomm® SA7255P	QNN_DLC	45.721 ms	2 - 260 MB	NPU	TrOCR.dlc
TrOCREncoder	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	TFLITE	12.102 ms	7 - 10 MB	NPU	TrOCR.tflite
TrOCREncoder	float	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_DLC	16.77 ms	2 - 4 MB	NPU	TrOCR.dlc
TrOCREncoder	float	SA8295P ADP	Qualcomm® SA8295P	TFLITE	20.681 ms	5 - 256 MB	NPU	TrOCR.tflite
TrOCREncoder	float	SA8295P ADP	Qualcomm® SA8295P	QNN_DLC	25.815 ms	2 - 262 MB	NPU	TrOCR.dlc
TrOCREncoder	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	TFLITE	12.098 ms	7 - 9 MB	NPU	TrOCR.tflite
TrOCREncoder	float	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_DLC	16.769 ms	2 - 4 MB	NPU	TrOCR.dlc
TrOCREncoder	float	SA8775P ADP	Qualcomm® SA8775P	TFLITE	14.633 ms	7 - 252 MB	NPU	TrOCR.tflite
TrOCREncoder	float	SA8775P ADP	Qualcomm® SA8775P	QNN_DLC	19.802 ms	2 - 263 MB	NPU	TrOCR.dlc
TrOCREncoder	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	TFLITE	8.523 ms	6 - 343 MB	NPU	TrOCR.tflite
TrOCREncoder	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_DLC	11.519 ms	2 - 356 MB	NPU	TrOCR.dlc
TrOCREncoder	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	ONNX	11.547 ms	16 - 390 MB	NPU	TrOCR.onnx.zip
TrOCREncoder	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	TFLITE	5.789 ms	6 - 253 MB	NPU	TrOCR.tflite
TrOCREncoder	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_DLC	9.302 ms	2 - 287 MB	NPU	TrOCR.dlc
TrOCREncoder	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	ONNX	10.061 ms	16 - 279 MB	NPU	TrOCR.onnx.zip
TrOCREncoder	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	TFLITE	4.642 ms	0 - 250 MB	NPU	TrOCR.tflite
TrOCREncoder	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_DLC	6.944 ms	2 - 265 MB	NPU	TrOCR.dlc
TrOCREncoder	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	ONNX	7.25 ms	16 - 259 MB	NPU	TrOCR.onnx.zip
TrOCREncoder	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_DLC	17.633 ms	2 - 2 MB	NPU	TrOCR.dlc
TrOCREncoder	float	Snapdragon X Elite CRD	Snapdragon® X Elite	ONNX	17.988 ms	48 - 48 MB	NPU	TrOCR.onnx.zip

Installation

Install the package via pip:

# NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported.
pip install "qai-hub-models[trocr]"

Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device

Sign-in to Qualcomm® AI Hub Workbench with your Qualcomm® ID. Once signed in navigate to Account -> Settings -> API Token.

With this API token, you can configure your client to run models on the cloud hosted devices.

qai-hub configure --api_token API_TOKEN

Navigate to docs for more information.

Demo off target

The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input.

python -m qai_hub_models.models.trocr.demo

The above demo runs a reference implementation of pre-processing, model inference, and post processing.

NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).

%run -m qai_hub_models.models.trocr.demo

Run model on a cloud-hosted device

In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following:

Performance check on-device on a cloud-hosted device
Downloads compiled assets that can be deployed on-device for Android.
Accuracy check between PyTorch and on-device outputs.

python -m qai_hub_models.models.trocr.export

How does this work?

This export script leverages Qualcomm® AI Hub to optimize, validate, and deploy this model on-device. Lets go through each step below in detail:

Step 1: Compile model for on-device deployment

To compile a PyTorch model for on-device deployment, we first trace the model in memory using the jit.trace and then call the submit_compile_job API.

import torch

import qai_hub as hub
from qai_hub_models.models.trocr import Model

# Load the model
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S25")

# Trace model
input_shape = torch_model.get_input_spec()
sample_inputs = torch_model.sample_inputs()

pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])

# Compile model on a specific device
compile_job = hub.submit_compile_job(
    model=pt_model,
    device=device,
    input_specs=torch_model.get_input_spec(),
)

# Get target model to run on-device
target_model = compile_job.get_target_model()

Step 2: Performance profiling on cloud-hosted device

After compiling models from step 1. Models can be profiled model on-device using the target_model. Note that this scripts runs the model on a device automatically provisioned in the cloud. Once the job is submitted, you can navigate to a provided job URL to view a variety of on-device performance metrics.

profile_job = hub.submit_profile_job(
    model=target_model,
    device=device,
)

Step 3: Verify on-device accuracy

To verify the accuracy of the model on-device, you can run on-device inference on sample input data on the same cloud hosted device.

input_data = torch_model.sample_inputs()
inference_job = hub.submit_inference_job(
    model=target_model,
    device=device,
    inputs=input_data,
)
    on_device_output = inference_job.download_output_data()

With the output of the model, you can compute like PSNR, relative errors or spot check the output with expected output.

Note: This on-device profiling and inference requires access to Qualcomm® AI Hub Workbench. Sign up for access.

Deploying compiled model to Android

The models can be deployed using multiple runtimes:

TensorFlow Lite (.tflite export): This tutorial provides a guide to deploy the .tflite model in an Android application.
QNN (.so export ): This sample app provides instructions on how to use the .so shared library in an Android application.

View on Qualcomm® AI Hub

Get more details on TrOCR's performance across various devices here. Explore all available models on Qualcomm® AI Hub

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: 515

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 • 11