### Build MRZScanner Wheel Package

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

After cloning the repository, build the wheel package for installation. This involves installing the wheel package and then running the build command.

```bash
pip install wheel
cd MRZScanner
python setup.py bdist_wheel
```

--------------------------------

### Install MRZScanner from Wheel File

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Install the MRZScanner library from the generated wheel file. Replace the asterisk with the actual version number.

```bash
pip install dist/mrzscanner_docsaid-*-py3-none-any.whl
```

--------------------------------

### Install MRZScanner via PyPI

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Install the mrzscanner-docsaid package using pip. This is the recommended method for most users.

```bash
pip install mrzscanner-docsaid
```

--------------------------------

### Configure MRZFinetuneDataset for Fine-tuning

Source: https://github.com/docsaidlab/mrzscanner/blob/main/dataset/README.md

Configure the MRZFinetuneDataset for the fine-tuning stage. This setup is similar to pre-training but omits the dataset length parameter.

```python
ds = MRZFinetuneDataset(
    root='/data/Dataset',
    image_size=(512, 512),
    aug_ratio=1.0,
    return_tensor=True,
)
```

--------------------------------

### Verify MRZScanner Installation

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Check if the mrzscanner library is installed correctly by printing its version number. A successful output indicates the installation was successful.

```python
import mrzscanner; print(mrzscanner.__version__)
```

--------------------------------

### Initialize MRZScanner with CUDA Backend

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Use this to initialize the MRZScanner with the CUDA backend for GPU computation. Ensure CUDA is installed and supported.

```python
from capybara import Backend

model = MRZScanner(backend=Backend.cuda) # Use CUDA backend
```

--------------------------------

### MRZScanner Spotting Model Inference Example

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Example of using the 'spotting' model type for end-to-end MRZ detection and recognition. This model does not return MRZ coordinates. Ensure necessary libraries like cv2 and skimage are installed.

```python
import cv2
from skimage import io
from mrzscanner import MRZScanner, ModelType

# Create model
model = MRZScanner(
   model_type=ModelType.spotting,
   spotting_cfg='20240919'
)

# Read image from the web
img = io.imread('https://github.com/DocsaidLab/MRZScanner/blob/main/docs/test_mrz.jpg?raw=true')
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)

# Model inference
result = model(img, do_center_crop=True, do_postprocess=False)

# Output result
print(result)
# {
#    'mrz_polygon': None,
#    'mrz_texts': [
#        'PCAZEQAOARIN<<FIDAN<<<<<<<<<<<<<<<<<<<<<<<<<',
#        'C946302620AZE6707297F23031072W12IMJ<<<<<<<40'
#    ],
#    'msg': <ErrorCodes.NO_ERROR: 'No error.'>
# }
```

--------------------------------

### Process Ground Truth String

Source: https://github.com/docsaidlab/mrzscanner/blob/main/dataset/README.md

The ground truth `gt` is a string, typically separated by '&'. Split this string to get individual MRZ codes, useful for text recognition tasks.

```python
print('GT original:', gt)
# >>> GT original: 9SZ563DNUUM1NEAKVHUI09MK4H9SAW&56QI04IS244AF4F0809416QQQ9MDOA&DS0CJGRAALNCT1NKPYGH5DPLAOY72L

print('GT:', gt.split('&'))
# >>> GT: ['9SZ563DNUUM1NEAKVHUI09MK4H9SAW',
#          '56QI04IS244AF4F0809416QQQ9MDOA',
#          'DS0CJGRAALNCT1NKPYGH5DPLAOY72L']
```

--------------------------------

### Initialize MRZ Finetune Dataset

Source: https://github.com/docsaidlab/mrzscanner/blob/main/dataset/README.md

Instantiate the `MRZFinetuneDataset` for fine-tuning. Ensure the `root` path points to your downloaded MIDV-2020 dataset. Setting `return_tensor=True` converts the output to PyTorch Tensors.

```python
from finetune_dataset import MRZFinetuneDataset

ds = MRZFinetuneDataset(
    root='/data/Dataset',  # Change to your dataset path
    image_size=(512, 512),
    aug_ratio=1.0,
    return_tensor=True,  # Converts the output to a PyTorch Tensor
)
```

--------------------------------

### Initialize MRZScanner with CPU Backend

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Use this to initialize the MRZScanner with the CPU backend for computation.

```python
from capybara import Backend

model = MRZScanner(backend=Backend.cpu) # Use CPU backend
```

--------------------------------

### Initialize MRZScanner with Spotting Model Configuration

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Initialize the MRZScanner using the 'spotting' model type and specifying the '20240919' configuration for spotting.

```python
model = MRZScanner(
   model_type=ModelType.spotting,
   spotting_cfg='20240919'
)
```

--------------------------------

### Configure MRZDataset for Pre-training

Source: https://github.com/docsaidlab/mrzscanner/blob/main/dataset/README.md

Use this configuration to set up the MRZDataset for the initial pre-training phase. It specifies image size, augmentation ratio, tensor return, and dataset length.

```python
ds = MRZDataset(
    root='/data/Dataset',
    image_size=(512, 512),
    aug_ratio=1.0,
    return_tensor=True,
    length_of_dataset=640000,
)
```

--------------------------------

### Access Fine-tuning Dataset Sample

Source: https://github.com/docsaidlab/mrzscanner/blob/main/dataset/README.md

Retrieve a single data sample (image and ground truth label) from the fine-tuning dataset. The dataset returns only the image and its corresponding MRZ text.

```python
img, gt = ds[1]
```

--------------------------------

### Initialize MRZScanner with Recognition Model Configuration

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Initialize the MRZScanner to load only the MRZ recognition model, using the '20250221' configuration.

```python
model = MRZScanner(
   model_type=ModelType.recognition,
   recognition_cfg='20250221'
)
```

--------------------------------

### Initialize MRZScanner with Detection Model Configuration

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Initialize the MRZScanner to load only the MRZ detection model, using the '20250222' configuration.

```python
model = MRZScanner(
   model_type=ModelType.detection,
   detection_cfg='20250222'
)
```

--------------------------------

### Initialize MRZDataset

Source: https://github.com/docsaidlab/mrzscanner/blob/main/dataset/README.md

Initializes the MRZDataset for generating synthetic MRZ images. Ensure external datasets are downloaded and paths are correctly set. The `return_tensor` option converts output to PyTorch Tensors.

```python
import cv2
from train_dataset import MRZDataset

ds = MRZDataset(
    root='/data/Dataset', # Change to your dataset path
    image_size=(512, 512),
    aug_ratio=1.0,
    return_tensor=True,  # Converts the output to a PyTorch Tensor
    length_of_dataset=1000000,  # Specify the size of the dataset
)
```

--------------------------------

### Initialize MRZScanner with Two-Stage Model Configuration

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Initialize the MRZScanner for a two-stage process, specifying configurations for both detection ('20250222') and recognition ('20250221').

```python
model = MRZScanner(
   model_type=ModelType.two_stage,
   detection_cfg='20250222',
   recognition_cfg='20250221'
)
```

--------------------------------

### Clone MRZScanner Repository

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Download the MRZScanner project from GitHub using git clone. This is useful for development or if you need the latest unreleased code.

```bash
git clone https://github.com/DocsaidLab/MRZScanner.git
```

--------------------------------

### Initialize MRZScanner with Spotting Model Type

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Initialize the MRZScanner to use the 'spotting' model type, which is an end-to-end model for detection and recognition.

```python
from mrzscanner import MRZScanner

model = MRZScanner(model_type=MRZScanner.spotting)
```

--------------------------------

### List Available MRZScanner Models

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Call this method to view all available model versions for 'spotting', 'detection', and 'recognition'.

```python
from mrzscanner import MRZScanner

print(MRZScanner().list_models())
# {
#    'spotting': ['20240919'],
#    'detection': ['20250222'],
#    'recognition': ['20250221']
# }
```

--------------------------------

### Prepare Cropped MRZ Image for Recognition

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

This snippet demonstrates how to prepare a cropped MRZ image using detected polygon coordinates. It utilizes `imwarp_quadrangle` for image transformation.

```python
import numpy as np
from skimage import io
from capybara import imwarp_quadrangle, imwrite

polygon = np.array([
    [ 158.536 , 1916.3734],
    [1682.7792, 1976.1683],
    [1677.1018, 2120.8926],
    [ 152.8586, 2061.0977]
], dtype=np.float32)

img = io.imread('https://github.com/DocsaidLab/MRZScanner/blob/main/docs/test_mrz.jpg?raw=true')
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)

mrz_img = imwarp_quadrangle(img, polygon)
imwrite(mrz_img)
```

--------------------------------

### MRZ Recognition Only

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Use the recognition model to perform MRZ text extraction on a pre-cropped MRZ image. Set `do_center_crop=False` as cropping is already done. Requires only the recognition configuration.

```python
from mrzscanner import MRZScanner, ModelType

# Create model
model = MRZScanner(
   model_type=ModelType.recognition,
   recognition_cfg='20250221'
)

# Input the cropped MRZ image
result = model(mrz_img, do_center_crop=False)

# Output result
print(result)
# { 
#     'mrz_polygon': None,
#     'mrz_texts': [
#         'PCAZEQAQARIN<<FIDAN<<<<<<<<<<<<<<<<<<<<<<<<<',
#         'C946302620AZE6707297F23031072W12IMJ<<<<<<<40'
#     ],
#     'msg': <ErrorCodes.NO_ERROR: 'No error.'>
# }
```

--------------------------------

### Print and Split Ground Truth String

Source: https://github.com/docsaidlab/mrzscanner/blob/main/dataset/README.md

Display the raw ground truth string and then split it by the '&' delimiter to separate different parts of the MRZ data. This is useful for inspecting the structure of the labels.

```python
print('GT:', gt)
# >>> GT: PCAZEABDULLAYEV<<DIL<<<<<<<<<<<<<<<<<<<<<<<<C193895647AZE9408148M28081525188L2V<<<<<<<42

print('GT:', gt.split('&'))
# >>> GT: ['PCAZEABDULLAYEV<<DIL<<<<<<<<<<<<<<<<<<<<<<<<',
#          'C193895647AZE9408148M28081525188L2V<<<<<<<42']
```

--------------------------------

### Use Center Cropping with MRZScanner

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Enable center cropping for images with non-standard aspect ratios to improve inference accuracy. This is useful for images taken with mobile devices or other sources that may not be square.

```python
from mrzscanner import MRZScanner

model = MRZScanner()

result = model(img, do_center_crop=True) # Use center cropping
```

--------------------------------

### Enable Post-processing with MRZScanner

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Enable post-processing to standardize MRZ blocks according to known rules, such as character case and valid characters for specific fields. This can help correct erroneous recognition results.

```python
result = model(img, do_postprocess=True)
```

--------------------------------

### MRZScanner Model Inference

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Perform MRZ (Machine Readable Zone) scanning on an image. The MRZScanner instance can be called directly for inference. The model automatically downloads if not found.

```python
import cv2
from skimage import io
from mrzscanner import MRZScanner

# Create model
model = MRZScanner()

# Read image from the web
img = io.imread('https://github.com/DocsaidLab/MRZScanner/blob/main/docs/test_mrz.jpg?raw=true')
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)

# Model inference
result = model(img, do_center_crop=True, do_postprocess=False)

# Output result
print(result)
```

--------------------------------

### Access MRZDataset Outputs

Source: https://github.com/docsaidlab/mrzscanner/blob/main/dataset/README.md

Retrieves a single data sample from the MRZDataset. The dataset generates data randomly, so results will vary even with the same index.

```python
img, gt, poly, fixed_points, mrz_points_hmap, mrz_region_hmap = ds[0]
```

--------------------------------

### Two-Stage MRZ Detection and Recognition

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Use the two_stage model for high accuracy MRZ detection and recognition. It returns MRZ coordinates and can be used to draw the detected location on the image. Requires both detection and recognition configurations.

```python
import cv2
from skimage import io
from mrzscanner import MRZScanner, ModelType

# Create model
model = MRZScanner(
   model_type=ModelType.two_stage,
   detection_cfg='20250222',
   recognition_cfg='20250221'
)

# Read image from the web
img = io.imread('https://github.com/DocsaidLab/MRZScanner/blob/main/docs/test_mrz.jpg?raw=true')
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)

# Model inference
result = model(img, do_center_crop=True, do_postprocess=False)

# Output result
print(result)
# { 
#     'mrz_polygon': 
#         array( 
#             [
#                 [ 158.536 , 1916.3734],
#                 [1682.7792, 1976.1683],
#                 [1677.1018, 2120.8926],
#                 [ 152.8586, 2061.0977]
#             ],
#             dtype=float32
#         ),
#     'mrz_texts': [
#         'PCAZEQAQARIN<<FIDAN<<<<<<<<<<<<<<<<<<<<<<<<<',
#         'C946302620AZE6707297F23031072W12IMJ<<<<<<<40'
#     ],
#     'msg': <ErrorCodes.NO_ERROR: 'No error.'>
# }

# Draw MRZ location
from capybara import draw_polygon, imwrite, centercrop

poly_img = draw_polygon(img, result['mrz_polygon'], color=(0, 0, 255), thickness=5)
imwrite(centercrop(poly_img))
```

--------------------------------

### Plot Character Center Points

Source: https://github.com/docsaidlab/mrzscanner/blob/main/dataset/README.md

Draws circles on the image at the center of each character within the MRZ, based on `fixed_points`. This is useful for precise character point localization tasks.

```python
import capybara as cb

point_img = img.copy()
for p in fixed_points:
    if p[0] == 0 and p[1] == 0:
        break
    cv2.circle(point_img, (int(p[0]), int(p[1])), 2, (0, 0, 255), -1)

cb.imwrite(point_img, 'points.jpg')
```

--------------------------------

### MRZScanner Citation

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

BibTeX entry for citing the MRZScanner project. Use this in academic papers and projects.

```bibtex
@misc{yuan2024mrzscanner,
  author = {Ze Yuan},
  title = {MRZScanner},
  year = {2024},
  publisher = {GitHub},
  url = {https://github.com/DocsaidLab/MRZScanner},
  note = {GitHub repository}
}
```

--------------------------------

### MRZ Detection Only

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

Use the detection model to only find the MRZ location without performing recognition. This model requires only the detection configuration.

```python
import cv2
from skimage import io
from mrzscanner import MRZScanner, ModelType

# Create model
model = MRZScanner(
   model_type=ModelType.detection,
   detection_cfg='20250222',
)

# Read image from the web
img = io.imread('https://github.com/DocsaidLab/MRZScanner/blob/main/docs/test_mrz.jpg?raw=true')
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)

# Model inference
result = model(img, do_center_crop=True)

# Output result
print(result)
# { 
#     'mrz_polygon': 
#         array( 
#             [
#                 [ 158.536 , 1916.3734],
#                 [1682.7792, 1976.1683],
#                 [1677.1018, 2120.8926],
#                 [ 152.8586, 2061.0977]
#             ],
#             dtype=float32
#         ),
#     'mrz_texts': None,
#     'msg': <ErrorCodes.NO_ERROR: 'No error.'>
# }
```

--------------------------------

### Draw MRZ Polygon

Source: https://github.com/docsaidlab/mrzscanner/blob/main/dataset/README.md

Visualizes the four points defining the MRZ area using `capybara.draw_polygon`. This is useful for region localization models.

```python
import capybara as cb

poly_img = cb.draw_polygon(img.copy(), poly, color=(0, 255, 0))
cb.imwrite(poly_img, 'poly.jpg')
```

--------------------------------

### MRZ Post-processing Function

Source: https://github.com/docsaidlab/mrzscanner/blob/main/README.md

A helper function to replace digits with visually similar characters that are valid in MRZ fields. This is part of the post-processing step to correct potential recognition errors.

```python
import re

def replace_digits(text: str):
    text = re.sub('0', 'O', text)
    text = re.sub('1', 'I', text)
    text = re.sub('2', 'Z', text)
    text = re.sub('4', 'A', text)
    text = re.sub('5', 'S', text)
    text = re.sub('8', 'B', text)
    return text

if doc_type == 3:  # TD1
    if len(results[0]) != 30 or len(results[1]) != 30 or len(results[2]) != 30:
        return [''], ErrorCodes.POSTPROCESS_FAILED_TD1_LENGTH
    # Line1
    doc = results[0][0:2]
    country = replace_digits(results[0][2:5])
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.