Method __init__

modelscope/preprocessors/multi_modal.py:449–490

Preprocess data for the model `VLDocForDocVLEmbedding`. Args: model_dir (str): model path in model hub. mode (str): model mode, in ('train', 'eval', 'inference').

(self,
                 model_dir: str,
                 mode: str = ModeKeys.INFERENCE,
                 *args,
                 **kwargs)

class VLDocPreprocessor(Preprocessor):

    def __init__(self,
                 model_dir: str,
                 mode: str = ModeKeys.INFERENCE,
                 *args,
                 **kwargs):
        """Preprocess data for the model `VLDocForDocVLEmbedding`.

        Args:
            model_dir (str): model path in model hub.
            mode (str): model mode, in ('train', 'eval', 'inference').
        """
        super().__init__(*args, **kwargs)

        self.model_dir = model_dir
        self.mode = mode

        model_cfg_path = osp.join(model_dir, 'config.json')
        with open(model_cfg_path, 'r', encoding='utf-8') as f:
            model_cfg = json.load(f)

        from modelscope.models.multi_modal.vldoc.tokenization import VLDocXLMTokenizer
        tokenizer_path = osp.join(model_dir, ModelFile.TOKENIZER_FOLDER)
        self.tokenizer = VLDocXLMTokenizer.from_pretrained(tokenizer_path)

        from modelscope.models.multi_modal.vldoc.processing import Processor, ImageProcessor
        self.img_proc = ImageProcessor(
            do_preprocess=True,
            do_resize=True,
            image_size={
                'height': model_cfg['image_size'][0],
                'width': model_cfg['image_size'][1],
            },
            do_normalize=True,
            apply_ocr=False)
        self.proc = Processor(
            max_seq_length=model_cfg['max_seq_length'],
            max_block_num=model_cfg['max_block_num'],
            img_processor=self.img_proc,
            tokenizer=self.tokenizer,
            width=model_cfg['image_size'][1],
            height=model_cfg['image_size'][0],
        )

    def __call__(self, input: Dict[str, Any], *args,
                 **kwargs) -> Dict[str, Any]: