bi*dZddlmZddlZddlmZddlmZmZm Z ddl Z ddl m Z ddlmcmcmZddlmZmZddlmZmZdd lmZmZd gZGd d eZy) zTransforms for text rendering and augmentation on images. This module provides transforms for adding and manipulating text on images, including text augmentation techniques like word insertion, deletion, and swapping. ) annotationsN)Path) AnnotatedAnyLiteral)AfterValidator) check_bboxesdenormalize_bboxes)check_range_bounds nondecreasing)BaseTransformInitSchemaImageOnlyTransform TextImageceZdZdZGddeZ d d fd Zed dZ ddZ ddZ ddZ dd Z dfd Z xZS)raApply text rendering transformations on images. This class supports rendering text directly onto images using a variety of configurations, such as custom fonts, font sizes, colors, and augmentation methods. The text can be placed inside specified bounding boxes. Args: font_path (str | Path): Path to the font file to use for rendering text. stopwords (list[str] | None): List of stopwords for text augmentation. augmentations (tuple[str | None, ...]): List of text augmentations to apply. None: text is printed as is "insertion": insert random stop words into the text. "swap": swap random words in the text. "deletion": delete random words from the text. fraction_range (tuple[float, float]): Range for selecting a fraction of bounding boxes to modify. font_size_fraction_range (tuple[float, float]): Range for selecting the font size as a fraction of bounding box height. font_color (tuple[float, ...]): Font color as RGB values (e.g., (0, 0, 0) for black). clear_bg (bool): Whether to clear the background before rendering text. metadata_key (str): Key to access metadata in the parameters. p (float): Probability of applying the transform. Targets: image, volume Image types: uint8, float32 References: doc-augmentation: https://github.com/danaaubakirova/doc-augmentation Examples: >>> import albumentations as A >>> transform = A.Compose([ A.TextImage( font_path=Path("/path/to/font.ttf"), stopwords=("the", "is", "in"), augmentations=("insertion", "deletion"), fraction_range=(0.5, 1.0), font_size_fraction_range=(0.5, 0.9), font_color=(255, 0, 0), # red in RGB metadata_key="text_metadata", p=0.5 ) ]) >>> transformed = transform(image=my_image, text_metadata=my_metadata) >>> image = transformed['image'] # This will render text on `my_image` based on the metadata provided in `my_metadata`. c^eZdZUded<ded<ded<ded<ded <d ed <d ed <ded<y)TextImage.InitSchema str | Path font_pathtuple[str, ...] stopwordsztuple[str | None, ...] augmentationszgAnnotated[tuple[float, float], AfterValidator(nondecreasing), AfterValidator(check_range_bounds(0, 1))]fraction_rangefont_size_fraction_rangetuple[float, ...] font_colorboolclear_bgstr metadata_keyN)__name__ __module__ __qualname____annotations__g/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/albumentations/augmentations/text/transforms.py InitSchemarLs;""--  #  &%r%r'c t || ||_||_||_||_t ||_||_||_ ||_ y)N)p) super__init__rrrrlistrrrr) selfrrrrrrrrr) __class__s r&r+zTextImage.__init__^sV 1(","!-0(@%$  r%c|jgS)zGet list of targets that should be passed as parameters to transforms. Returns: list[str]: List containing the metadata key name )r)r-s r&targets_as_paramszTextImage.targets_as_paramsts!!""r%c |jjDcgc]}|s| }}t|}tdt ||z}|dk(r-t j |||j|j}nY|dk(r"t j|||j}n2|dk(r"t j|||j}n tdtjdd|j}||k7r|SdScc}w) aApply a random text augmentation to the input text. Args: text (str): Original text to augment fraction (float): Fraction of words to modify choice (Literal["insertion", "swap", "deletion"]): Type of augmentation to apply Returns: str: Augmented text or empty string if no change was made Raises: ValueError: If an invalid choice is provided  insertionswapdeletionz?Invalid choice. Choose from 'insertion', 'swap', or 'deletion'.z + )stripsplitlenmaxintftextinsert_random_stopwordsr py_randomswap_random_wordsdelete_random_words ValueErrorresub) r-textfractionchoicewordwords num_wordsnum_words_to_modifyresult_sentences r& random_augzTextImage.random_aug~s(#'**,"4"4"6?$$??J !!SI)=%>? [ #;;ECVX\XfXfhlhvhvwO v #55e=PRVR`R`aO z !#77?RTXTbTbcO^_ _&&sO<BBD"1T"9ArA@s D D c ddlm}tt j |gt t j |g|jddd}d|ddD\}} } } | | z } |jj|j} |jt|jt| | z}|jr |j|}n=|jj!|j}||n|j#|d| }|j$}|| | | f|||||d S#t$r}td|d}~wwxYw) aPreprocess text metadata for a single bounding box. Args: image (np.ndarray): Input image bbox (tuple[float, float, float, float]): Normalized bounding box coordinates text (str): Text to render in the bounding box bbox_index (int): Index of the bounding box in the original metadata Returns: dict[str, Any]: Processed metadata including font, position, and text information Raises: ImportError: If PIL.ImageFont is not installed r) ImageFontz`ImageFont from PIL is required to use TextImage transform. Install it with `pip install Pillow`.Nc32K|]}t|ywN)r<).0xs r& z0TextImage.preprocess_metadata..s%Lc!f%Ls?)rG) bbox_coords bbox_index original_textrEfontr)PILrO ImportErrorr nparrayr shaper?uniformrtruetyperrr<rrGrMr)r-imagebboxrErYrOerrdenormalized_bboxx_miny_minx_maxy_max bbox_heightfont_size_fractionr[augmented_text augmentationrs r&preprocess_metadatazTextImage.preprocess_metadatasT,  % RXXtf%&.rxx/?RaQRST%L6G6K%L"ueUem 3T^^33T5R5RS!!#dnn"5s;MP[;[7\]!!T%7%7%?!N>>001C1CDL%1%9TttUXam?nN__ "5%7$!"$   / r  sD22 E ; EE c d|vr|dn|dd}||j}|gk(rdgiSt|tr|g}|jj|j }t t||z}|jjtt||}|Dcgc]"}|j|||d||d|$} }d| iScc}w)aLGenerate parameters based on input data. Args: params (dict[str, Any]): Dictionary of existing parameters data (dict[str, Any]): Dictionary containing input data with image and metadata Returns: dict[str, Any]: Dictionary containing the overlay data for text rendering rcimagesr overlay_datardrE) r isinstancedictr?rarr<r:samplerangero) r-paramsdatarcmetadatarFnum_bboxes_to_modifybbox_indices_to_updaterYrrs r&get_params_dependent_on_dataz&TextImage.get_params_dependent_on_datas")DW d8nQ6G))* r>  h % zH)4>>))4+>+>?"3x=8#;<!%!6!6uS]7KMa!b5   $ $UHZ,@,H(S]J^_eJfhr s  L   s)'Cc Ftj|||jS)aLApply text rendering to the input image. Args: img (np.ndarray): Input image overlay_data (list[dict[str, Any]]): List of dictionaries containing text rendering information **params (Any): Additional parameters Returns: np.ndarray: Image with rendered text )r)r= render_textr)r-imgrrrws r&applyzTextImage.applys"  lT]]KKr%c t||g|i|}|dDcgc]}|d|d|d|d|ddc}|d<|Scc}w)aApply the transform and include overlay data in the result. Args: params (dict[str, Any]): Parameters for the transform *args (Any): Additional positional arguments **kwargs (Any): Additional keyword arguments Returns: dict[str, Any]: Dictionary containing the transformed data and simplified overlay information rrrXrErZrYr)rXrErZrYr)r*apply_with_params)r-rwargskwargsresoverlayr.s r&rzTextImage.apply_with_paramssyg'@@@".1   '}5!(!9%l3%l3    N   sA))theisinatofrR)?r)g?g?)rrrFtextimage_metadatarW)rrrrrz;tuple[Literal['insertion', 'swap', 'deletion'] | None, ...]rtuple[float, float]rrrrrrrrr)floatreturnNone)rz list[str])rErrFrrGz(Literal['insertion', 'swap', 'deletion']rr) rc np.ndarrayrdz!tuple[float, float, float, float]rErrYr<rdict[str, Any])rwrrxrrr)rrrrzlist[dict[str, Any]]rwrrr)rwrrrrrrr)r r!r"__doc__r r'r+propertyr0rMror|rr __classcell__)r.s@r&rrsT1f,*&EU\.88B(10!!#!S ! , ! #6 !&!!! ! !,##"B"B"B9 "B "BH6 6 06  6  6  6 p$ LL L+L L  L&r%)r __future__rrCpathlibrtypingrrrnumpyr^pydanticr,albumentations.augmentations.text.functionalrrE functionalr=albumentations.core.bbox_utilsr r albumentations.core.pydanticr r (albumentations.core.transforms_interfacer r__all__rr$r%r&rsF # **#<<KJ` -S"Sr%