Recognizers

Text recognition models.

class manuscript.recognizers.TRBA(weights=None, config=None, charset=None, device=None, force_download=False, rotate_threshold=1.5, region_preparer='bbox', region_preparer_options=None, min_text_size=5, batch_size=128, **kwargs)[source]

Bases: BaseRecognizer

Инициализация модели распознавания текста TRBA с использованием ONNX Runtime.

Methods

__call__(*args, **kwargs)

Call self as a function.

export(weights_path, config_path, ...[, ...])

Экспорт модели TRBA PyTorch в формат ONNX.

predict(page[, image, batch_size, ...])

Распознаёт текст для текстовых областей на Page и возвращает обновлённый Page.

runtime_providers()

Get ONNX Runtime execution providers based on device.

train(train_csvs, train_roots[, val_csvs, ...])

Обучение модели распознавания текста TRBA на пользовательских наборах данных.
Parameters:

  • weights (str | None)

  • config (str | None)

  • charset (str | None)

  • device (str | None)

  • force_download (bool)

  • rotate_threshold (float | None)

  • region_preparer (str | Callable[[...], Sequence[Any]])

  • region_preparer_options (Dict[str, Any] | None)

  • min_text_size (int)

  • batch_size (int)

__init__(weights=None, config=None, charset=None, device=None, force_download=False, rotate_threshold=1.5, region_preparer='bbox', region_preparer_options=None, min_text_size=5, batch_size=128, **kwargs)[source]
Parameters:

  • weights (str | None)

  • config (str | None)

  • charset (str | None)

  • device (str | None)

  • force_download (bool)

  • rotate_threshold (float | None)

  • region_preparer (str | Callable[[...], Sequence[Any]])

  • region_preparer_options (Dict[str, Any] | None)

  • min_text_size (int)

  • batch_size (int)

charset_registry = {'trba_base_g1': 'github://konstantinkozhin/manuscript-ocr/v0.1.0/trba_base_g1.txt', 'trba_lite_g1': 'github://konstantinkozhin/manuscript-ocr/v0.1.0/trba_lite_g1.txt', 'trba_lite_g2': 'github://konstantinkozhin/manuscript-ocr/v0.1.0/trba_lite_g2.txt'}
config_registry = {'trba_base_g1': 'github://konstantinkozhin/manuscript-ocr/v0.1.0/trba_base_g1.json', 'trba_lite_g1': 'github://konstantinkozhin/manuscript-ocr/v0.1.0/trba_lite_g1.json', 'trba_lite_g2': 'github://konstantinkozhin/manuscript-ocr/v0.1.0/trba_lite_g2.json'}
default_weights_name: Optional[str] = 'trba_lite_g1'
static export(weights_path, config_path, charset_path, output_path, opset_version=14, simplify=True)[source]

Экспорт модели TRBA PyTorch в формат ONNX.

Метод конвертирует обученную модель TRBA из PyTorch в формат ONNX, который может использоваться для более быстрого инференса с ONNX Runtime. Экспортированную модель можно загрузить через TRBA(weights="model.onnx").

Return type:

None

Parameters:

  • weights_path (str | Path)

  • config_path (str | Path)

  • charset_path (str | Path)

  • output_path (str | Path)

  • opset_version (int)

  • simplify (bool)

predict(page, image=None, batch_size=None, debug_save_dir=None)[source]

Распознаёт текст для текстовых областей на Page и возвращает обновлённый Page.

Return type:

Page

Parameters:

  • page (Page)

  • image (numpy.ndarray | str | Path | Image | None)

  • batch_size (int | None)

  • debug_save_dir (str | Path | None)

pretrained_registry: Dict[str, str] = {'trba_base_g1': 'github://konstantinkozhin/manuscript-ocr/v0.1.0/trba_base_g1.onnx', 'trba_lite_g1': 'github://konstantinkozhin/manuscript-ocr/v0.1.0/trba_lite_g1.onnx', 'trba_lite_g2': 'github://konstantinkozhin/manuscript-ocr/v0.1.0/trba_lite_g2.onnx'}
static train(train_csvs, train_roots, val_csvs=None, val_roots=None, *, exp_dir=None, charset_path=None, encoding='utf-8', img_h=64, img_w=256, max_len=25, hidden_size=256, num_encoder_layers=3, cnn_in_channels=3, cnn_out_channels=512, cnn_backbone='seresnet31', ctc_weight=0.3, ctc_weight_decay_epochs=50, ctc_weight_min=0.0, max_grad_norm=5.0, batch_size=32, epochs=20, lr=0.001, optimizer='AdamW', scheduler='OneCycleLR', weight_decay=0.0, momentum=0.9, val_interval=1, val_size=3000, train_proportions=None, num_workers=0, seed=42, resume_from=None, save_interval=None, device='cuda', freeze_cnn='none', freeze_enc_rnn='none', freeze_attention='none', pretrain_weights='default', **extra_config)[source]

Обучение модели распознавания текста TRBA на пользовательских наборах данных.

Parameters:

  • train_csvs (str | Sequence[str])

  • train_roots (str | Sequence[str])

  • val_csvs (str | Sequence[str] | None)

  • val_roots (str | Sequence[str] | None)

  • exp_dir (str | None)

  • charset_path (str | None)

  • encoding (str)

  • img_h (int)

  • img_w (int)

  • max_len (int)

  • hidden_size (int)

  • num_encoder_layers (int)

  • cnn_in_channels (int)

  • cnn_out_channels (int)

  • cnn_backbone (str)

  • ctc_weight (float)

  • ctc_weight_decay_epochs (int)

  • ctc_weight_min (float)

  • max_grad_norm (float)

  • batch_size (int)

  • epochs (int)

  • lr (float)

  • optimizer (str)

  • scheduler (str)

  • weight_decay (float)

  • momentum (float)

  • val_interval (int)

  • val_size (int)

  • train_proportions (Sequence[float] | None)

  • num_workers (int)

  • seed (int)

  • resume_from (str | None)

  • save_interval (int | None)

  • device (str)

  • freeze_cnn (str)

  • freeze_enc_rnn (str)

  • freeze_attention (str)

  • pretrain_weights (object | None)

  • extra_config (Any)