PaddleX/paddlex/inference/models/text_recognition/predictor.py

# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from ....modules.text_recognition.model_list import MODELS
from ....utils.func_register import FuncRegister
from ...common.batch_sampler import ImageBatchSampler
from ...common.reader import ReadImage
from ..base import BasePredictor
from .processors import CTCLabelDecode, OCRReisizeNormImg, ToBatch
from .result import TextRecResult


class TextRecPredictor(BasePredictor):

    entities = MODELS

    _FUNC_MAP = {}
    register = FuncRegister(_FUNC_MAP)

    def __init__(self, *args, input_shape=None, **kwargs):
        super().__init__(*args, **kwargs)
        self.input_shape = input_shape
        self.pre_tfs, self.infer, self.post_op = self._build()

    def _build_batch_sampler(self):
        return ImageBatchSampler()

    def _get_result_class(self):
        return TextRecResult

    def _build(self):
        pre_tfs = {"Read": ReadImage(format="RGB")}
        for cfg in self.config["PreProcess"]["transform_ops"]:
            tf_key = list(cfg.keys())[0]
            assert tf_key in self._FUNC_MAP
            func = self._FUNC_MAP[tf_key]
            args = cfg.get(tf_key, {})
            name, op = func(self, **args) if args else func(self)
            if op:
                pre_tfs[name] = op
        pre_tfs["ToBatch"] = ToBatch()

        infer = self.create_static_infer()

        post_op = self.build_postprocess(**self.config["PostProcess"])
        return pre_tfs, infer, post_op

    def process(self, batch_data):
        batch_raw_imgs = self.pre_tfs["Read"](imgs=batch_data.instances)
        batch_imgs = self.pre_tfs["ReisizeNorm"](imgs=batch_raw_imgs)
        x = self.pre_tfs["ToBatch"](imgs=batch_imgs)
        batch_preds = self.infer(x=x)
        texts, scores = self.post_op(batch_preds)
        return {
            "input_path": batch_data.input_paths,
            "page_index": batch_data.page_indexes,
            "input_img": batch_raw_imgs,
            "rec_text": texts,
            "rec_score": scores,
        }

    @register("DecodeImage")
    def build_readimg(self, channel_first, img_mode):
        assert channel_first == False
        return "Read", ReadImage(format=img_mode)

    @register("RecResizeImg")
    def build_resize(self, image_shape):
        return "ReisizeNorm", OCRReisizeNormImg(
            rec_image_shape=image_shape, input_shape=self.input_shape
        )

    def build_postprocess(self, **kwargs):
        if kwargs.get("name") == "CTCLabelDecode":
            return CTCLabelDecode(
                character_list=kwargs.get("character_dict"),
            )
        else:
            raise Exception()

    @register("MultiLabelEncode")
    def foo(self, *args, **kwargs):
        return None, None

    @register("KeepKeys")
    def foo(self, *args, **kwargs):
        return None, None