Preprocess

apply_pipeline(images, pipeline)

Applies a pipeline (i.e. list of transformations)

Parameters:

Name	Type	Description	Default
images	list	List of images to be transformed	required
pipeline	list	List of transformation to be applied	required

Returns: list: Preprocessed images

Source code in template_vision/preprocessing/preprocess.py

def apply_pipeline(images: list, pipeline: list) -> list:
    '''Applies a pipeline (i.e. list of transformations)

    Args:
        images (list): List of images to be transformed
        pipeline (list): List of transformation to be applied
    Returns:
        list: Preprocessed images
    '''
    # Process
    results = None
    for transformer in pipeline:
        if results is None:
            results = transformer(images)
        else:
            results = transformer(results)
    return results

convert_rgb(images)

Converts a list of image into RGB images

Parameters:

Name	Type	Description	Default
images	list	List of images to be converted	required

Returns: list: RGB images

Source code in template_vision/preprocessing/preprocess.py

def convert_rgb(images: list) -> list:
    '''Converts a list of image into RGB images

    Args:
        images (list): List of images to be converted
    Returns:
        list: RGB images
    '''
    return [im.convert('RGB') for im in images]

get_preprocessor(preprocess_str)

Gets a preprocessing (function) from its name

Parameters:

Name	Type	Description	Default
preprocess_str	str	Name of the preprocess	required

Raises: ValueError: If the name of the preprocess is not known Returns: Callable: Function to be used for the preprocessing

Source code in template_vision/preprocessing/preprocess.py

def get_preprocessor(preprocess_str: str) -> Callable:
    '''Gets a preprocessing (function) from its name

    Args:
        preprocess_str (str): Name of the preprocess
    Raises:
        ValueError: If the name of the preprocess is not known
    Returns:
        Callable: Function to be used for the preprocessing
    '''
    # Process
    preprocessors_dict = get_preprocessors_dict()
    if preprocess_str not in preprocessors_dict.keys():
        raise ValueError(f"The preprocess {preprocess_str} is not known.")
    # Get preprocessor
    preprocessor = preprocessors_dict[preprocess_str]
    # Return
    return preprocessor

get_preprocessors_dict()

Gets a dictionary of available preprocessing

Returns:

Name	Type	Description
dict	dict	Dictionary of preprocessing

Source code in template_vision/preprocessing/preprocess.py

def get_preprocessors_dict() -> dict:
    '''Gets a dictionary of available preprocessing

    Returns:
        dict: Dictionary of preprocessing
    '''
    preprocessors_dict = {
        'no_preprocess': lambda x: x,  # - /!\ DO NOT DELETE -> necessary for compatibility /!\ -
        'preprocess_convert_rgb': preprocess_convert_rgb,  # Simple RGB converter
        'preprocess_docs': preprocess_docs,  # Example pipeline with documents (remove white borders, 3/4 ratio, resize 224x224)
    }
    return preprocessors_dict

jpeg_compression(images, quality=75)

Simulates a JPEG compression Might be useful for prediction if a model is trained with JPEG compressed images.

Parameters:

Name	Type	Description	Default
images	list	List of images to be compressed	required

Kwargs: quality (int): Wanted quality Returns: list: Compressed images

Source code in template_vision/preprocessing/preprocess.py

def jpeg_compression(images: list, quality: int = 75) -> list:
    '''Simulates a JPEG compression
    Might be useful for prediction if a model is trained with JPEG compressed images.

    Args:
        images (list): List of images to be compressed
    Kwargs:
        quality (int): Wanted quality
    Returns:
        list: Compressed images
    '''
    # Process images one by one
    results = []
    for i, im in enumerate(tqdm.tqdm(images)):
        out = BytesIO()
        im.save(out, format='JPEG', quality=quality)
        out.seek(0)
        tmp_im = Image.open(out)
        tmp_im.load()  # Cumpulsory to avoid "Too many open files" issue
        results.append(tmp_im)
    return results

preprocess_convert_rgb(images)

Applies a simple RGB conversion

Parameters:

Name	Type	Description	Default
images	list	List of images to be transformed	required

Returns: list: Preprocessed images

Source code in template_vision/preprocessing/preprocess.py

def preprocess_convert_rgb(images: list) -> list:
    '''Applies a simple RGB conversion

    Args:
        images (list): List of images to be transformed
    Returns:
        list: Preprocessed images
    '''
    # Process
    pipeline = [convert_rgb]
    return apply_pipeline(images, pipeline=pipeline)

preprocess_docs(images)

Applies a list of usual transformations with scanned documents

Parameters:

Name	Type	Description	Default
images	list	List of images to be transformed	required

Returns: list: Preprocessed images

Source code in template_vision/preprocessing/preprocess.py

def preprocess_docs(images: list) -> list:
    '''Applies a list of usual transformations with scanned documents

    Args:
        images (list): List of images to be transformed
    Returns:
        list: Preprocessed images
    '''
    # Process
    pipeline = [convert_rgb, functools.partial(manage_white_borders.remove_white_borders, image_ratio_strategy='fill', image_ratio=0.75, with_rotation=True), resize]
    return apply_pipeline(images, pipeline=pipeline)

resize(images, width=224, height=224)

Resizes images

Parameters:

Name	Type	Description	Default
images	list	List of images to be resized	required

Kwargs: width (int): Wanted width height (int): Wanted height Raises: ValueError: If width < 1 ValueError: If height < 1 Returns: list: Resized images

Source code in template_vision/preprocessing/preprocess.py

def resize(images: list, width: int = 224, height: int = 224) -> list:
    '''Resizes images

    Args:
        images (list): List of images to be resized
    Kwargs:
        width (int): Wanted width
        height (int): Wanted height
    Raises:
        ValueError: If width < 1
        ValueError: If height < 1
    Returns:
        list: Resized images
    '''
    if width < 1:
        raise ValueError('Width must be strictly positive.')
    if height < 1:
        raise ValueError('Height must be strictly positive.')
    # Process images one by one
    results = []
    for i, im in enumerate(tqdm.tqdm(images)):
        results.append(im.resize((width, height)))
    return results