README.md

---
library_name: pytorch
license: other
tags:
- android
pipeline_tag: image-to-text

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/trocr/web-assets/model_demo.png)

# 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](https://huggingface.co/microsoft/trocr-small-stage1).


This repository provides scripts to run TrOCR on Qualcomm® devices.
More details on model performance across various devices, can be found
[here](https://aihub.qualcomm.com/models/trocr).



### 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 | 4.222 ms | 0 - 72 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 4.002 ms | 5 - 67 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 2.632 ms | 0 - 132 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 3.018 ms | 4 - 132 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 2.066 ms | 0 - 214 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 2.01 ms | 1 - 27 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 2.69 ms | 0 - 179 MB | NPU | [TrOCR.onnx.zip](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.onnx.zip) |
| TrOCRDecoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 2.886 ms | 0 - 72 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 2.813 ms | 5 - 67 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 4.222 ms | 0 - 72 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 4.002 ms | 5 - 67 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 2.081 ms | 0 - 215 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 2.011 ms | 2 - 28 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 3.157 ms | 0 - 64 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 2.877 ms | 0 - 59 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 2.06 ms | 0 - 224 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 2.032 ms | 2 - 26 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 2.886 ms | 0 - 72 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 2.813 ms | 5 - 67 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 2.114 ms | 0 - 244 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 2.014 ms | 1 - 27 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 2.677 ms | 0 - 179 MB | NPU | [TrOCR.onnx.zip](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.onnx.zip) |
| TrOCRDecoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 1.541 ms | 0 - 148 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 1.443 ms | 0 - 139 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 1.829 ms | 0 - 139 MB | NPU | [TrOCR.onnx.zip](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.onnx.zip) |
| TrOCRDecoder | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 1.399 ms | 0 - 73 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCRDecoder | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 1.269 ms | 2 - 152 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 1.66 ms | 1 - 150 MB | NPU | [TrOCR.onnx.zip](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.onnx.zip) |
| TrOCRDecoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 2.219 ms | 626 - 626 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCRDecoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 2.471 ms | 68 - 68 MB | NPU | [TrOCR.onnx.zip](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.onnx.zip) |
| TrOCREncoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 48.039 ms | 7 - 115 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 46.257 ms | 2 - 139 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 25.412 ms | 7 - 128 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 28.129 ms | 1 - 130 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 17.742 ms | 7 - 23 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 17.229 ms | 2 - 28 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 19.204 ms | 0 - 135 MB | NPU | [TrOCR.onnx.zip](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.onnx.zip) |
| TrOCREncoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 20.329 ms | 3 - 111 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 19.824 ms | 2 - 137 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 48.039 ms | 7 - 115 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 46.257 ms | 2 - 139 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 17.733 ms | 7 - 26 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 17.299 ms | 2 - 21 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 28.15 ms | 7 - 119 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 25.233 ms | 2 - 126 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 17.794 ms | 7 - 24 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 17.31 ms | 2 - 24 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 20.329 ms | 3 - 111 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 19.824 ms | 2 - 137 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 17.804 ms | 7 - 25 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 17.254 ms | 2 - 25 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 19.718 ms | 0 - 134 MB | NPU | [TrOCR.onnx.zip](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.onnx.zip) |
| TrOCREncoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 12.3 ms | 6 - 117 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 11.66 ms | 2 - 143 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 14.315 ms | 15 - 159 MB | NPU | [TrOCR.onnx.zip](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.onnx.zip) |
| TrOCREncoder | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 10.719 ms | 6 - 118 MB | NPU | [TrOCR.tflite](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.tflite) |
| TrOCREncoder | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 9.067 ms | 2 - 157 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 13.236 ms | 14 - 152 MB | NPU | [TrOCR.onnx.zip](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.onnx.zip) |
| TrOCREncoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 18.037 ms | 190 - 190 MB | NPU | [TrOCR.dlc](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.dlc) |
| TrOCREncoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 19.153 ms | 49 - 49 MB | NPU | [TrOCR.onnx.zip](https://huggingface.co/qualcomm/TrOCR/blob/main/TrOCR.onnx.zip) |




## Installation


Install the package via pip:
```bash
pip install "qai-hub-models[trocr]"
```


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

Sign-in to [Qualcomm® AI Hub](https://app.aihub.qualcomm.com/) 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.
```bash
qai-hub configure --api_token API_TOKEN
```
Navigate to [docs](https://app.aihub.qualcomm.com/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.

```bash
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.

```bash
python -m qai_hub_models.models.trocr.export
```



## How does this work?

This [export script](https://aihub.qualcomm.com/models/trocr/qai_hub_models/models/TrOCR/export.py)
leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) 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.

```python
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 S24")

# 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.
```python
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.
```python
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. [Sign up for access](https://myaccount.qualcomm.com/signup).




## Deploying compiled model to Android


The models can be deployed using multiple runtimes:
- TensorFlow Lite (`.tflite` export): [This
  tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a
  guide to deploy the .tflite model in an Android application.


- QNN (`.so` export ): This [sample
  app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html)
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](https://aihub.qualcomm.com/models/trocr).
Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)


## License
* The license for the original implementation of TrOCR can be found
  [here](https://github.com/microsoft/unilm/blob/master/LICENSE).
* The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf)



## References
* [TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models](https://arxiv.org/abs/2109.10282)
* [Source Model Implementation](https://huggingface.co/microsoft/trocr-small-stage1)



## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).