paddlex 3.0.0b2__py3-none-any.whl → 3.0.0rc1__py3-none-any.whl

paddlex/inference/pipelines/ocr/pipeline.py

@@ -0,0 +1,406 @@
+# 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 typing import Any, Dict, List, Optional, Union
+
+import numpy as np
+
+from ....utils import logging
+from ....utils.deps import pipeline_requires_extra
+from ...common.batch_sampler import ImageBatchSampler
+from ...common.reader import ReadImage
+from ...utils.hpi import HPIConfig
+from ...utils.pp_option import PaddlePredictorOption
+from ..base import BasePipeline
+from ..components import (
+    CropByPolys,
+    SortPolyBoxes,
+    SortQuadBoxes,
+    convert_points_to_boxes,
+    rotate_image,
+)
+from .result import OCRResult
+
+
+@pipeline_requires_extra("ocr")
+class OCRPipeline(BasePipeline):
+    """OCR Pipeline"""
+
+    entities = "OCR"
+
+    def __init__(
+        self,
+        config: Dict,
+        device: Optional[str] = None,
+        pp_option: Optional[PaddlePredictorOption] = None,
+        use_hpip: bool = False,
+        hpi_config: Optional[Union[Dict[str, Any], HPIConfig]] = None,
+    ) -> None:
+        """
+        Initializes the class with given configurations and options.
+
+        Args:
+            config (Dict): Configuration dictionary containing various settings.
+            device (str, optional): Device to run the predictions on. Defaults to None.
+            pp_option (PaddlePredictorOption, optional): PaddlePredictor options. Defaults to None.
+            use_hpip (bool, optional): Whether to use the high-performance
+                inference plugin (HPIP). Defaults to False.
+            hpi_config (Optional[Union[Dict[str, Any], HPIConfig]], optional):
+                The high-performance inference configuration dictionary.
+                Defaults to None.
+        """
+        super().__init__(
+            device=device, pp_option=pp_option, use_hpip=use_hpip, hpi_config=hpi_config
+        )
+
+        self.use_doc_preprocessor = config.get("use_doc_preprocessor", True)
+        if self.use_doc_preprocessor:
+            doc_preprocessor_config = config.get("SubPipelines", {}).get(
+                "DocPreprocessor",
+                {
+                    "pipeline_config_error": "config error for doc_preprocessor_pipeline!"
+                },
+            )
+            self.doc_preprocessor_pipeline = self.create_pipeline(
+                doc_preprocessor_config
+            )
+
+        self.use_textline_orientation = config.get("use_textline_orientation", True)
+        if self.use_textline_orientation:
+            textline_orientation_config = config.get("SubModules", {}).get(
+                "TextLineOrientation",
+                {"model_config_error": "config error for textline_orientation_model!"},
+            )
+            self.textline_orientation_model = self.create_model(
+                textline_orientation_config
+            )
+
+        text_det_config = config.get("SubModules", {}).get(
+            "TextDetection", {"model_config_error": "config error for text_det_model!"}
+        )
+        self.text_type = config["text_type"]
+        if self.text_type == "general":
+            self.text_det_limit_side_len = text_det_config.get("limit_side_len", 960)
+            self.text_det_limit_type = text_det_config.get("limit_type", "max")
+            self.text_det_thresh = text_det_config.get("thresh", 0.3)
+            self.text_det_box_thresh = text_det_config.get("box_thresh", 0.6)
+            self.input_shape = text_det_config.get("input_shape", None)
+            self.text_det_unclip_ratio = text_det_config.get("unclip_ratio", 2.0)
+            self._sort_boxes = SortQuadBoxes()
+            self._crop_by_polys = CropByPolys(det_box_type="quad")
+        elif self.text_type == "seal":
+            self.text_det_limit_side_len = text_det_config.get("limit_side_len", 736)
+            self.text_det_limit_type = text_det_config.get("limit_type", "min")
+            self.text_det_thresh = text_det_config.get("thresh", 0.2)
+            self.text_det_box_thresh = text_det_config.get("box_thresh", 0.6)
+            self.text_det_unclip_ratio = text_det_config.get("unclip_ratio", 0.5)
+            self.input_shape = text_det_config.get("input_shape", None)
+            self._sort_boxes = SortPolyBoxes()
+            self._crop_by_polys = CropByPolys(det_box_type="poly")
+        else:
+            raise ValueError("Unsupported text type {}".format(self.text_type))
+
+        self.text_det_model = self.create_model(
+            text_det_config,
+            limit_side_len=self.text_det_limit_side_len,
+            limit_type=self.text_det_limit_type,
+            thresh=self.text_det_thresh,
+            box_thresh=self.text_det_box_thresh,
+            unclip_ratio=self.text_det_unclip_ratio,
+            input_shape=self.input_shape,
+        )
+
+        text_rec_config = config.get("SubModules", {}).get(
+            "TextRecognition",
+            {"model_config_error": "config error for text_rec_model!"},
+        )
+        self.text_rec_score_thresh = text_rec_config.get("score_thresh", 0)
+        self.input_shape = text_rec_config.get("input_shape", None)
+        self.text_rec_model = self.create_model(
+            text_rec_config, input_shape=self.input_shape
+        )
+
+        self.batch_sampler = ImageBatchSampler(batch_size=1)
+        self.img_reader = ReadImage(format="BGR")
+
+    def rotate_image(
+        self, image_array_list: List[np.ndarray], rotate_angle_list: List[int]
+    ) -> List[np.ndarray]:
+        """
+        Rotate the given image arrays by their corresponding angles.
+        0 corresponds to 0 degrees, 1 corresponds to 180 degrees.
+
+        Args:
+            image_array_list (List[np.ndarray]): A list of input image arrays to be rotated.
+            rotate_angle_list (List[int]): A list of rotation indicators (0 or 1).
+                                        0 means rotate by 0 degrees
+                                        1 means rotate by 180 degrees
+
+        Returns:
+            List[np.ndarray]: A list of rotated image arrays.
+
+        Raises:
+            AssertionError: If any rotate_angle is not 0 or 1.
+            AssertionError: If the lengths of input lists don't match.
+        """
+        assert len(image_array_list) == len(
+            rotate_angle_list
+        ), f"Length of image_array_list ({len(image_array_list)}) must match length of rotate_angle_list ({len(rotate_angle_list)})"
+
+        for angle in rotate_angle_list:
+            assert angle in [0, 1], f"rotate_angle must be 0 or 1, now it's {angle}"
+
+        rotated_images = []
+        for image_array, rotate_indicator in zip(image_array_list, rotate_angle_list):
+            # Convert 0/1 indicator to actual rotation angle
+            rotate_angle = rotate_indicator * 180
+            rotated_image = rotate_image(image_array, rotate_angle)
+            rotated_images.append(rotated_image)
+
+        return rotated_images
+
+    def check_model_settings_valid(self, model_settings: Dict) -> bool:
+        """
+        Check if the input parameters are valid based on the initialized models.
+
+        Args:
+            model_info_params(Dict): A dictionary containing input parameters.
+
+        Returns:
+            bool: True if all required models are initialized according to input parameters, False otherwise.
+        """
+
+        if model_settings["use_doc_preprocessor"] and not self.use_doc_preprocessor:
+            logging.error(
+                "Set use_doc_preprocessor, but the models for doc preprocessor are not initialized."
+            )
+            return False
+
+        if (
+            model_settings["use_textline_orientation"]
+            and not self.use_textline_orientation
+        ):
+            logging.error(
+                "Set use_textline_orientation, but the models for use_textline_orientation are not initialized."
+            )
+            return False
+
+        return True
+
+    def get_model_settings(
+        self,
+        use_doc_orientation_classify: Optional[bool],
+        use_doc_unwarping: Optional[bool],
+        use_textline_orientation: Optional[bool],
+    ) -> dict:
+        """
+        Get the model settings based on the provided parameters or default values.
+
+        Args:
+            use_doc_orientation_classify (Optional[bool]): Whether to use document orientation classification.
+            use_doc_unwarping (Optional[bool]): Whether to use document unwarping.
+            use_textline_orientation (Optional[bool]): Whether to use textline orientation.
+
+        Returns:
+            dict: A dictionary containing the model settings.
+        """
+        if use_doc_orientation_classify is None and use_doc_unwarping is None:
+            use_doc_preprocessor = self.use_doc_preprocessor
+        else:
+            if use_doc_orientation_classify is True or use_doc_unwarping is True:
+                use_doc_preprocessor = True
+            else:
+                use_doc_preprocessor = False
+
+        if use_textline_orientation is None:
+            use_textline_orientation = self.use_textline_orientation
+        return dict(
+            use_doc_preprocessor=use_doc_preprocessor,
+            use_textline_orientation=use_textline_orientation,
+        )
+
+    def get_text_det_params(
+        self,
+        text_det_limit_side_len: Optional[int] = None,
+        text_det_limit_type: Optional[str] = None,
+        text_det_thresh: Optional[float] = None,
+        text_det_box_thresh: Optional[float] = None,
+        text_det_unclip_ratio: Optional[float] = None,
+    ) -> dict:
+        """
+        Get text detection parameters.
+
+        If a parameter is None, its default value from the instance will be used.
+
+        Args:
+            text_det_limit_side_len (Optional[int]): The maximum side length of the text box.
+            text_det_limit_type (Optional[str]): The type of limit to apply to the text box.
+            text_det_thresh (Optional[float]): The threshold for text detection.
+            text_det_box_thresh (Optional[float]): The threshold for the bounding box.
+            text_det_unclip_ratio (Optional[float]): The ratio for unclipping the text box.
+
+        Returns:
+            dict: A dictionary containing the text detection parameters.
+        """
+        if text_det_limit_side_len is None:
+            text_det_limit_side_len = self.text_det_limit_side_len
+        if text_det_limit_type is None:
+            text_det_limit_type = self.text_det_limit_type
+        if text_det_thresh is None:
+            text_det_thresh = self.text_det_thresh
+        if text_det_box_thresh is None:
+            text_det_box_thresh = self.text_det_box_thresh
+        if text_det_unclip_ratio is None:
+            text_det_unclip_ratio = self.text_det_unclip_ratio
+        return dict(
+            limit_side_len=text_det_limit_side_len,
+            limit_type=text_det_limit_type,
+            thresh=text_det_thresh,
+            box_thresh=text_det_box_thresh,
+            unclip_ratio=text_det_unclip_ratio,
+        )
+
+    def predict(
+        self,
+        input: Union[str, List[str], np.ndarray, List[np.ndarray]],
+        use_doc_orientation_classify: Optional[bool] = None,
+        use_doc_unwarping: Optional[bool] = None,
+        use_textline_orientation: Optional[bool] = None,
+        text_det_limit_side_len: Optional[int] = None,
+        text_det_limit_type: Optional[str] = None,
+        text_det_thresh: Optional[float] = None,
+        text_det_box_thresh: Optional[float] = None,
+        text_det_unclip_ratio: Optional[float] = None,
+        text_rec_score_thresh: Optional[float] = None,
+    ) -> OCRResult:
+        """
+        Predict OCR results based on input images or arrays with optional preprocessing steps.
+
+        Args:
+            input (Union[str, list[str], np.ndarray, list[np.ndarray]]): Input image of pdf path(s) or numpy array(s).
+            use_doc_orientation_classify (Optional[bool]): Whether to use document orientation classification.
+            use_doc_unwarping (Optional[bool]): Whether to use document unwarping.
+            use_textline_orientation (Optional[bool]): Whether to use textline orientation prediction.
+            text_det_limit_side_len (Optional[int]): Maximum side length for text detection.
+            text_det_limit_type (Optional[str]): Type of limit to apply for text detection.
+            text_det_thresh (Optional[float]): Threshold for text detection.
+            text_det_box_thresh (Optional[float]): Threshold for text detection boxes.
+            text_det_unclip_ratio (Optional[float]): Ratio for unclipping text detection boxes.
+            text_rec_score_thresh (Optional[float]): Score threshold for text recognition.
+        Returns:
+            OCRResult: Generator yielding OCR results for each input image.
+        """
+
+        model_settings = self.get_model_settings(
+            use_doc_orientation_classify, use_doc_unwarping, use_textline_orientation
+        )
+
+        if not self.check_model_settings_valid(model_settings):
+            yield {"error": "the input params for model settings are invalid!"}
+
+        text_det_params = self.get_text_det_params(
+            text_det_limit_side_len,
+            text_det_limit_type,
+            text_det_thresh,
+            text_det_box_thresh,
+            text_det_unclip_ratio,
+        )
+
+        if text_rec_score_thresh is None:
+            text_rec_score_thresh = self.text_rec_score_thresh
+
+        for img_id, batch_data in enumerate(self.batch_sampler(input)):
+            image_array = self.img_reader(batch_data.instances)[0]
+
+            if model_settings["use_doc_preprocessor"]:
+                doc_preprocessor_res = next(
+                    self.doc_preprocessor_pipeline(
+                        image_array,
+                        use_doc_orientation_classify=use_doc_orientation_classify,
+                        use_doc_unwarping=use_doc_unwarping,
+                    )
+                )
+            else:
+                doc_preprocessor_res = {"output_img": image_array}
+
+            doc_preprocessor_image = doc_preprocessor_res["output_img"]
+
+            det_res = next(
+                self.text_det_model(doc_preprocessor_image, **text_det_params)
+            )
+
+            dt_polys = det_res["dt_polys"]
+            det_res["dt_scores"]
+
+            dt_polys = self._sort_boxes(dt_polys)
+
+            single_img_res = {
+                "input_path": batch_data.input_paths[0],
+                "page_index": batch_data.page_indexes[0],
+                "doc_preprocessor_res": doc_preprocessor_res,
+                "dt_polys": dt_polys,
+                "model_settings": model_settings,
+                "text_det_params": text_det_params,
+                "text_type": self.text_type,
+                "text_rec_score_thresh": text_rec_score_thresh,
+            }
+
+            single_img_res["rec_texts"] = []
+            single_img_res["rec_scores"] = []
+            single_img_res["rec_polys"] = []
+            if len(dt_polys) > 0:
+                all_subs_of_img = list(
+                    self._crop_by_polys(doc_preprocessor_image, dt_polys)
+                )
+                # use textline orientation model
+                if model_settings["use_textline_orientation"]:
+                    angles = [
+                        int(textline_angle_info["class_ids"][0])
+                        for textline_angle_info in self.textline_orientation_model(
+                            all_subs_of_img
+                        )
+                    ]
+                    all_subs_of_img = self.rotate_image(all_subs_of_img, angles)
+                else:
+                    angles = [-1] * len(all_subs_of_img)
+                single_img_res["textline_orientation_angles"] = angles
+
+                sub_img_info_list = [
+                    {
+                        "sub_img_id": img_id,
+                        "sub_img_ratio": sub_img.shape[1] / float(sub_img.shape[0]),
+                    }
+                    for img_id, sub_img in enumerate(all_subs_of_img)
+                ]
+                sorted_subs_info = sorted(
+                    sub_img_info_list, key=lambda x: x["sub_img_ratio"]
+                )
+                sorted_subs_of_img = [
+                    all_subs_of_img[x["sub_img_id"]] for x in sorted_subs_info
+                ]
+                for idx, rec_res in enumerate(self.text_rec_model(sorted_subs_of_img)):
+                    sub_img_id = sorted_subs_info[idx]["sub_img_id"]
+                    sub_img_info_list[sub_img_id]["rec_res"] = rec_res
+                for sno in range(len(sub_img_info_list)):
+                    rec_res = sub_img_info_list[sno]["rec_res"]
+                    if rec_res["rec_score"] >= text_rec_score_thresh:
+                        single_img_res["rec_texts"].append(rec_res["rec_text"])
+                        single_img_res["rec_scores"].append(rec_res["rec_score"])
+                        single_img_res["rec_polys"].append(dt_polys[sno])
+            if self.text_type == "general":
+                rec_boxes = convert_points_to_boxes(single_img_res["rec_polys"])
+                single_img_res["rec_boxes"] = rec_boxes
+            else:
+                single_img_res["rec_boxes"] = np.array([])
+            yield OCRResult(single_img_res)

paddlex/inference/pipelines/ocr/result.py

@@ -0,0 +1,252 @@
+# 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.
+
+import math
+import random
+from pathlib import Path
+from typing import Dict
+
+import numpy as np
+from PIL import Image, ImageDraw
+
+from ....utils.deps import class_requires_deps, function_requires_deps, is_dep_available
+from ....utils.fonts import SIMFANG_FONT_FILE_PATH, create_font
+from ...common.result import BaseCVResult, JsonMixin
+
+if is_dep_available("opencv-contrib-python"):
+    import cv2
+
+
+@class_requires_deps("opencv-contrib-python")
+class OCRResult(BaseCVResult):
+    """OCR result"""
+
+    def _get_input_fn(self):
+        fn = super()._get_input_fn()
+        if (page_idx := self["page_index"]) is not None:
+            fp = Path(fn)
+            stem, suffix = fp.stem, fp.suffix
+            return f"{stem}_{page_idx}{suffix}"
+        else:
+            return fn
+
+    def get_minarea_rect(self, points: np.ndarray) -> np.ndarray:
+        """
+        Get the minimum area rectangle for the given points using OpenCV.
+
+        Args:
+            points (np.ndarray): An array of 2D points.
+
+        Returns:
+            np.ndarray: An array of 2D points representing the corners of the minimum area rectangle
+                     in a specific order (clockwise or counterclockwise starting from the top-left corner).
+        """
+        bounding_box = cv2.minAreaRect(points)
+        points = sorted(list(cv2.boxPoints(bounding_box)), key=lambda x: x[0])
+
+        index_a, index_b, index_c, index_d = 0, 1, 2, 3
+        if points[1][1] > points[0][1]:
+            index_a = 0
+            index_d = 1
+        else:
+            index_a = 1
+            index_d = 0
+        if points[3][1] > points[2][1]:
+            index_b = 2
+            index_c = 3
+        else:
+            index_b = 3
+            index_c = 2
+
+        box = np.array(
+            [points[index_a], points[index_b], points[index_c], points[index_d]]
+        ).astype(np.int32)
+
+        return box
+
+    def _to_img(self) -> Dict[str, Image.Image]:
+        """
+        Converts the internal data to a PIL Image with detection and recognition results.
+
+        Returns:
+            Dict[Image.Image]: A dictionary containing two images: 'doc_preprocessor_res' and 'ocr_res_img'.
+        """
+        boxes = self["rec_polys"]
+        txts = self["rec_texts"]
+        image = self["doc_preprocessor_res"]["output_img"]
+        h, w = image.shape[0:2]
+        image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
+        img_left = Image.fromarray(image_rgb)
+        img_right = np.ones((h, w, 3), dtype=np.uint8) * 255
+        random.seed(0)
+        draw_left = ImageDraw.Draw(img_left)
+        for idx, (box, txt) in enumerate(zip(boxes, txts)):
+            try:
+                color = (
+                    random.randint(0, 255),
+                    random.randint(0, 255),
+                    random.randint(0, 255),
+                )
+                box = np.array(box)
+                if len(box) > 4:
+                    pts = [(x, y) for x, y in box.tolist()]
+                    draw_left.polygon(pts, outline=color, width=8)
+                    box = self.get_minarea_rect(box)
+                    height = int(0.5 * (max(box[:, 1]) - min(box[:, 1])))
+                    box[:2, 1] = np.mean(box[:, 1])
+                    box[2:, 1] = np.mean(box[:, 1]) + min(20, height)
+                draw_left.polygon(box, fill=color)
+                img_right_text = draw_box_txt_fine(
+                    (w, h), box, txt, SIMFANG_FONT_FILE_PATH
+                )
+                pts = np.array(box, np.int32).reshape((-1, 1, 2))
+                cv2.polylines(img_right_text, [pts], True, color, 1)
+                img_right = cv2.bitwise_and(img_right, img_right_text)
+            except:
+                continue
+
+        img_left = Image.blend(Image.fromarray(image_rgb), img_left, 0.5)
+        img_show = Image.new("RGB", (w * 2, h), (255, 255, 255))
+        img_show.paste(img_left, (0, 0, w, h))
+        img_show.paste(Image.fromarray(img_right), (w, 0, w * 2, h))
+
+        model_settings = self["model_settings"]
+        res_img_dict = {f"ocr_res_img": img_show}
+        if model_settings["use_doc_preprocessor"]:
+            res_img_dict.update(**self["doc_preprocessor_res"].img)
+        return res_img_dict
+
+    def _to_str(self, *args, **kwargs) -> Dict[str, str]:
+        """Converts the instance's attributes to a dictionary and then to a string.
+
+        Args:
+            *args: Additional positional arguments passed to the base class method.
+            **kwargs: Additional keyword arguments passed to the base class method.
+
+        Returns:
+            Dict[str, str]: A dictionary with the instance's attributes converted to strings.
+        """
+        data = {}
+        data["input_path"] = self["input_path"]
+        data["page_index"] = self["page_index"]
+        data["model_settings"] = self["model_settings"]
+        if self["model_settings"]["use_doc_preprocessor"]:
+            data["doc_preprocessor_res"] = self["doc_preprocessor_res"].str["res"]
+        data["dt_polys"] = (
+            self["dt_polys"]
+            if self["text_type"] == "seal"
+            else np.array(self["dt_polys"])
+        )
+        data["text_det_params"] = self["text_det_params"]
+        data["text_type"] = self["text_type"]
+        if "textline_orientation_angles" in self:
+            data["textline_orientation_angles"] = np.array(
+                self["textline_orientation_angles"]
+            )
+        data["text_rec_score_thresh"] = self["text_rec_score_thresh"]
+        data["rec_texts"] = self["rec_texts"]
+        data["rec_scores"] = np.array(self["rec_scores"])
+        data["rec_polys"] = (
+            self["rec_polys"]
+            if self["text_type"] == "seal"
+            else np.array(self["rec_polys"])
+        )
+        data["rec_boxes"] = np.array(self["rec_boxes"])
+
+        return JsonMixin._to_str(data, *args, **kwargs)
+
+    def _to_json(self, *args, **kwargs) -> Dict[str, str]:
+        """
+        Converts the object's data to a JSON dictionary.
+
+        Args:
+            *args: Positional arguments passed to the JsonMixin._to_json method.
+            **kwargs: Keyword arguments passed to the JsonMixin._to_json method.
+
+        Returns:
+            Dict[str, str]: A dictionary containing the object's data in JSON format.
+        """
+        data = {}
+        data["input_path"] = self["input_path"]
+        data["page_index"] = self["page_index"]
+        data["model_settings"] = self["model_settings"]
+        if self["model_settings"]["use_doc_preprocessor"]:
+            data["doc_preprocessor_res"] = self["doc_preprocessor_res"].json["res"]
+        data["dt_polys"] = self["dt_polys"]
+        data["text_det_params"] = self["text_det_params"]
+        data["text_type"] = self["text_type"]
+        if "textline_orientation_angles" in self:
+            data["textline_orientation_angles"] = self["textline_orientation_angles"]
+        data["text_rec_score_thresh"] = self["text_rec_score_thresh"]
+        data["rec_texts"] = self["rec_texts"]
+        data["rec_scores"] = self["rec_scores"]
+        data["rec_polys"] = self["rec_polys"]
+        data["rec_boxes"] = self["rec_boxes"]
+        return JsonMixin._to_json(data, *args, **kwargs)
+
+
+# Adds a function comment according to Google Style Guide
+@function_requires_deps("opencv-contrib-python")
+def draw_box_txt_fine(
+    img_size: tuple, box: np.ndarray, txt: str, font_path: str
+) -> np.ndarray:
+    """
+    Draws text in a box on an image with fine control over size and orientation.
+
+    Args:
+        img_size (tuple): The size of the output image (width, height).
+        box (np.ndarray): A 4x2 numpy array defining the corners of the box in (x, y) order.
+        txt (str): The text to draw inside the box.
+        font_path (str): The path to the font file to use for drawing the text.
+
+    Returns:
+        np.ndarray: An image with the text drawn in the specified box.
+    """
+    box_height = int(
+        math.sqrt((box[0][0] - box[3][0]) ** 2 + (box[0][1] - box[3][1]) ** 2)
+    )
+    box_width = int(
+        math.sqrt((box[0][0] - box[1][0]) ** 2 + (box[0][1] - box[1][1]) ** 2)
+    )
+
+    if box_height > 2 * box_width and box_height > 30:
+        img_text = Image.new("RGB", (box_height, box_width), (255, 255, 255))
+        draw_text = ImageDraw.Draw(img_text)
+        if txt:
+            font = create_font(txt, (box_height, box_width), font_path)
+            draw_text.text([0, 0], txt, fill=(0, 0, 0), font=font)
+        img_text = img_text.transpose(Image.ROTATE_270)
+    else:
+        img_text = Image.new("RGB", (box_width, box_height), (255, 255, 255))
+        draw_text = ImageDraw.Draw(img_text)
+        if txt:
+            font = create_font(txt, (box_width, box_height), font_path)
+            draw_text.text([0, 0], txt, fill=(0, 0, 0), font=font)
+
+    pts1 = np.float32(
+        [[0, 0], [box_width, 0], [box_width, box_height], [0, box_height]]
+    )
+    pts2 = np.array(box, dtype=np.float32)
+    M = cv2.getPerspectiveTransform(pts1, pts2)
+
+    img_text = np.array(img_text, dtype=np.uint8)
+    img_right_text = cv2.warpPerspective(
+        img_text,
+        M,
+        img_size,
+        flags=cv2.INTER_NEAREST,
+        borderMode=cv2.BORDER_CONSTANT,
+        borderValue=(255, 255, 255),
+    )
+    return img_right_text

paddlex/inference/pipelines/open_vocabulary_detection/__init__.py

@@ -0,0 +1,15 @@
+# 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 .pipeline import OpenVocabularyDetectionPipeline