Class Processor

modelscope/models/multi_modal/vldoc/processing.py:443–538

r"""Construct a GeoLayoutLM processor. Args: max_seq_length: max length for token max_block_num: max number of text lines (blocks or segments) img_processor: type of ImageProcessor. tokenizer: to tokenize strings. use_roberta_tokenizer: Whether the tokeni

class Processor(object):
    r"""Construct a GeoLayoutLM processor.

    Args:
        max_seq_length: max length for token
        max_block_num: max number of text lines (blocks or segments)
        img_processor: type of ImageProcessor.
        tokenizer: to tokenize strings.
        use_roberta_tokenizer: Whether the tokenizer is originated from RoBerta tokenizer
            (True by default).
        ocr_utils: a tool to preprocess ocr_infos.
        width: default width. It can be used only when all the images are of the same shape.
        height: default height. It can be used only when all the images are of the same shape.

    In `serialize_from_tokens`, the 3 sequences (i.e., `input_ids`, `bboxes_line`, `bboxes_word`)
        must not contain special tokens like [CLS] or [SEP].
    The boxes in `bboxes_line` and `bboxes_word` can be presented by either 2 points or 4 points.
    The value in boxes should keep original.
    Here is an example of the 3 arguments:
        ```
        input_ids ->
        [[6, 2391, 6, 31833, 6, 10132, 6, 2283, 6, 17730, 6, 2698, 152]]
        bboxes_line ->
        [[[230, 1, 353, 38], [230, 1, 353, 38], [230, 1, 353, 38], [230, 1, 353, 38],
            [230, 1, 353, 38], [230, 1, 353, 38], [230, 1, 353, 38], [230, 1, 353, 38],
            [257, 155, 338, 191], [257, 155, 338, 191], [257, 155, 338, 191], [257, 155, 338, 191],
            [257, 155, 338, 191]]]
        bboxes_word ->
        [[[231, 2, 267, 2, 267, 38, 231, 38], [231, 2, 267, 2, 267, 38, 231, 38],
            [264, 7, 298, 7, 298, 36, 264, 36], [264, 7, 298, 7, 298, 36, 264, 36],
            [293, 3, 329, 3, 329, 41, 293, 41], [293, 3, 329, 3, 329, 41, 293, 41],
            [330, 4, 354, 4, 354, 39, 330, 39], [330, 4, 354, 4, 354, 39, 330, 39],
            [258, 156, 289, 156, 289, 193, 258, 193], [258, 156, 289, 156, 289, 193, 258, 193],
            [288, 158, 321, 158, 321, 192, 288, 192], [288, 158, 321, 158, 321, 192, 288, 192],
            [321, 156, 336, 156, 336, 190, 321, 190]]]
        ```

    """

    def __init__(self,
                 max_seq_length,
                 max_block_num,
                 img_processor: ImageProcessor,
                 tokenizer=None,
                 use_roberta_tokenizer: bool = True,
                 ocr_utils: OCRUtils = None,
                 width=768,
                 height=768,
                 **kwargs):
        self.img_processor = img_processor
        self.tokenizer = tokenizer
        self.kwargs = kwargs

        self.serializer = TextLayoutSerializer(
            max_seq_length,
            max_block_num,
            tokenizer,
            width,