PyPI - paddlex - Versions diffs - 3.0.0rc1__py3-none-any.whl → 3.0.1__py3-none-any.whl - Mend

paddlex 3.0.0rc1py3-none-any.whl → 3.0.1py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (233) hide show

paddlex/inference/pipelines/layout_parsing/result_v2.py CHANGED Viewed

@@ -14,12 +14,15 @@
 from __future__ import annotations
 import copy
+import math
 import re
-from pathlib import Path
+from functools import partial
+from typing import List
 import numpy as np
-from PIL import Image, ImageDraw
+from PIL import Image, ImageDraw, ImageFont
+from ....utils.fonts import PINGFANG_FONT_FILE_PATH
 from ...common.result import (
     BaseCVResult,
     HtmlMixin,
@@ -27,7 +30,166 @@ from ...common.result import (
     MarkdownMixin,
     XlsxMixin,
 )
-from .utils import get_show_color
+from .setting import BLOCK_LABEL_MAP
+def compile_title_pattern():
+    # Precompiled regex pattern for matching numbering at the beginning of the title
+    numbering_pattern = (
+        r"(?:" + r"[1-9][0-9]*(?:\.[1-9][0-9]*)*[\.、]?|" + r"[\(\（](?:[1-9][0-9]*|["
+        r"一二三四五六七八九十百千万亿零壹贰叁肆伍陆柒捌玖拾]+)[\)\）]|" + r"["
+        r"一二三四五六七八九十百千万亿零壹贰叁肆伍陆柒捌玖拾]+"
+        r"[、\.]?|" + r"(?:I|II|III|IV|V|VI|VII|VIII|IX|X)\.?" + r")"
+    )
+    return re.compile(r"^\s*(" + numbering_pattern + r")(\s*)(.*)$")
+TITLE_RE_PATTERN = compile_title_pattern()
+def format_title_func(block):
+    """
+    Normalize chapter title.
+    Add the '#' to indicate the level of the title.
+    If numbering exists, ensure there's exactly one space between it and the title content.
+    If numbering does not exist, return the original title unchanged.
+    :param title: Original chapter title string.
+    :return: Normalized chapter title string.
+    """
+    title = block.content
+    match = TITLE_RE_PATTERN.match(title)
+    if match:
+        numbering = match.group(1).strip()
+        title_content = match.group(3).lstrip()
+        # Return numbering and title content separated by one space
+        title = numbering + " " + title_content
+    title = title.rstrip(".")
+    level = (
+        title.count(
+            ".",
+        )
+        + 1
+        if "." in title
+        else 1
+    )
+    return f"#{'#' * level} {title}".replace("-\n", "").replace(
+        "\n",
+        " ",
+    )
+def format_centered_by_html(string):
+    return (
+        f'<div style="text-align: center;">{string}</div>'.replace(
+            "-\n",
+            "",
+        ).replace("\n", " ")
+        + "\n"
+    )
+def format_text_plain_func(block):
+    return block.content
+def format_image_scaled_by_html_func(block, original_image_width):
+    img_tags = []
+    image_path = block.image["path"]
+    image_width = block.image["img"].width
+    scale = int(image_width / original_image_width * 100)
+    img_tags.append(
+        '<img src="{}" alt="Image" width="{}%" />'.format(
+            image_path.replace("-\n", "").replace("\n", " "), scale
+        ),
+    )
+    return "\n".join(img_tags)
+def format_image_plain_func(block):
+    img_tags = []
+    image_path = block.image["path"]
+    img_tags.append("![]({})".format(image_path.replace("-\n", "").replace("\n", " ")))
+    return "\n".join(img_tags)
+def format_chart2table_func(block):
+    lines_list = block.content.split("\n")
+    column_num = len(lines_list[0].split("|"))
+    lines_list.insert(1, "|".join(["---"] * column_num))
+    lines_list = [f"|{line}|" for line in lines_list]
+    return "\n".join(lines_list)
+def simplify_table_func(table_code):
+    return "\n" + table_code.replace("<html>", "").replace("</html>", "").replace(
+        "<body>", ""
+    ).replace("</body>", "")
+def format_first_line_func(block, templates, format_func, spliter):
+    lines = block.content.split(spliter)
+    for idx in range(len(lines)):
+        line = lines[idx]
+        if line.strip() == "":
+            continue
+        if line.lower() in templates:
+            lines[idx] = format_func(line)
+        break
+    return spliter.join(lines)
+def get_seg_flag(block: LayoutParsingBlock, prev_block: LayoutParsingBlock):
+    seg_start_flag = True
+    seg_end_flag = True
+    block_box = block.bbox
+    context_left_coordinate = block_box[0]
+    context_right_coordinate = block_box[2]
+    seg_start_coordinate = block.seg_start_coordinate
+    seg_end_coordinate = block.seg_end_coordinate
+    if prev_block is not None:
+        prev_block_bbox = prev_block.bbox
+        num_of_prev_lines = prev_block.num_of_lines
+        pre_block_seg_end_coordinate = prev_block.seg_end_coordinate
+        prev_end_space_small = (
+            abs(prev_block_bbox[2] - pre_block_seg_end_coordinate) < 10
+        )
+        prev_lines_more_than_one = num_of_prev_lines > 1
+        overlap_blocks = context_left_coordinate < prev_block_bbox[2]
+        # update context_left_coordinate and context_right_coordinate
+        if overlap_blocks:
+            context_left_coordinate = min(prev_block_bbox[0], context_left_coordinate)
+            context_right_coordinate = max(prev_block_bbox[2], context_right_coordinate)
+            prev_end_space_small = (
+                abs(context_right_coordinate - pre_block_seg_end_coordinate) < 10
+            )
+            edge_distance = 0
+        else:
+            edge_distance = abs(block_box[0] - prev_block_bbox[2])
+        current_start_space_small = seg_start_coordinate - context_left_coordinate < 10
+        if (
+            prev_end_space_small
+            and current_start_space_small
+            and prev_lines_more_than_one
+            and edge_distance < max(prev_block.width, block.width)
+        ):
+            seg_start_flag = False
+    else:
+        if seg_start_coordinate - context_left_coordinate < 10:
+            seg_start_flag = False
+    if context_right_coordinate - seg_end_coordinate < 10:
+        seg_end_flag = False
+    return seg_start_flag, seg_end_flag
 class LayoutParsingResultV2(BaseCVResult, HtmlMixin, XlsxMixin, MarkdownMixin):
@@ -40,30 +202,10 @@ class LayoutParsingResultV2(BaseCVResult, HtmlMixin, XlsxMixin, MarkdownMixin):
         XlsxMixin.__init__(self)
         MarkdownMixin.__init__(self)
         JsonMixin.__init__(self)
-        self.title_pattern = self._build_title_pattern()
-    def _build_title_pattern(self):
-        # Precompiled regex pattern for matching numbering at the beginning of the title
-        numbering_pattern = (
-            r"(?:"
-            + r"[1-9][0-9]*(?:\.[1-9][0-9]*)*[\.、]?|"
-            + r"[\(\（](?:[1-9][0-9]*|["
-            r"一二三四五六七八九十百千万亿零壹贰叁肆伍陆柒捌玖拾]+)[\)\）]|" + r"["
-            r"一二三四五六七八九十百千万亿零壹贰叁肆伍陆柒捌玖拾]+"
-            r"[、\.]?|" + r"(?:I|II|III|IV|V|VI|VII|VIII|IX|X)\.?" + r")"
-        )
-        return re.compile(r"^\s*(" + numbering_pattern + r")(\s*)(.*)$")
-    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 _to_img(self) -> dict[str, np.ndarray]:
+        from .utils import get_show_color
         res_img_dict = {}
         model_settings = self["model_settings"]
         if model_settings["use_doc_preprocessor"]:
@@ -71,12 +213,14 @@ class LayoutParsingResultV2(BaseCVResult, HtmlMixin, XlsxMixin, MarkdownMixin):
                 res_img_dict[key] = value
         res_img_dict["layout_det_res"] = self["layout_det_res"].img["res"]
-        if model_settings["use_general_ocr"] or model_settings["use_table_recognition"]:
-            res_img_dict["overall_ocr_res"] = self["overall_ocr_res"].img["ocr_res_img"]
+        if model_settings["use_region_detection"]:
+            res_img_dict["region_det_res"] = self["region_det_res"].img["res"]
+        res_img_dict["overall_ocr_res"] = self["overall_ocr_res"].img["ocr_res_img"]
         if model_settings["use_table_recognition"] and len(self["table_res_list"]) > 0:
             table_cell_img = Image.fromarray(
-                copy.deepcopy(self["doc_preprocessor_res"]["output_img"])
+                copy.deepcopy(self["doc_preprocessor_res"]["output_img"][:, :, ::-1])
             )
             table_draw = ImageDraw.Draw(table_cell_img)
             rectangle_color = (255, 0, 0)
@@ -101,16 +245,23 @@ class LayoutParsingResultV2(BaseCVResult, HtmlMixin, XlsxMixin, MarkdownMixin):
         # for layout ordering image
         image = Image.fromarray(self["doc_preprocessor_res"]["output_img"][:, :, ::-1])
         draw = ImageDraw.Draw(image, "RGBA")
-        parsing_result = self["parsing_res_list"]
+        font_size = int(0.018 * int(image.width)) + 2
+        font = ImageFont.truetype(PINGFANG_FONT_FILE_PATH, font_size, encoding="utf-8")
+        parsing_result: List[LayoutParsingBlock] = self["parsing_res_list"]
         for block in parsing_result:
-            bbox = block["block_bbox"]
-            index = block.get("index", None)
-            label = block["sub_label"]
-            fill_color = get_show_color(label)
+            bbox = block.bbox
+            index = block.order_index
+            label = block.label
+            fill_color = get_show_color(label, False)
             draw.rectangle(bbox, fill=fill_color)
             if index is not None:
-                text_position = (bbox[2] + 2, bbox[1] - 10)
-                draw.text(text_position, str(index), fill="red")
+                text_position = (bbox[2] + 2, bbox[1] - font_size // 2)
+                if int(image.width) - bbox[2] < font_size:
+                    text_position = (
+                        int(bbox[2] - font_size * 1.1),
+                        bbox[1] - font_size // 2,
+                    )
+                draw.text(text_position, str(index), font=font, fill="red")
         res_img_dict["layout_order_res"] = image
@@ -134,8 +285,7 @@ class LayoutParsingResultV2(BaseCVResult, HtmlMixin, XlsxMixin, MarkdownMixin):
         if self["model_settings"]["use_doc_preprocessor"]:
             data["doc_preprocessor_res"] = self["doc_preprocessor_res"].str["res"]
         data["layout_det_res"] = self["layout_det_res"].str["res"]
-        if model_settings["use_general_ocr"] or model_settings["use_table_recognition"]:
-            data["overall_ocr_res"] = self["overall_ocr_res"].str["res"]
+        data["overall_ocr_res"] = self["overall_ocr_res"].str["res"]
         if model_settings["use_table_recognition"] and len(self["table_res_list"]) > 0:
             data["table_res_list"] = []
             for sno in range(len(self["table_res_list"])):
@@ -176,9 +326,9 @@ class LayoutParsingResultV2(BaseCVResult, HtmlMixin, XlsxMixin, MarkdownMixin):
         parsing_res_list = self["parsing_res_list"]
         parsing_res_list = [
             {
-                "block_label": parsing_res["block_label"],
-                "block_content": parsing_res["block_content"],
-                "block_bbox": parsing_res["block_bbox"],
+                "block_label": parsing_res.label,
+                "block_content": parsing_res.content,
+                "block_bbox": parsing_res.bbox,
             }
             for parsing_res in parsing_res_list
         ]
@@ -186,8 +336,7 @@ class LayoutParsingResultV2(BaseCVResult, HtmlMixin, XlsxMixin, MarkdownMixin):
         if self["model_settings"]["use_doc_preprocessor"]:
             data["doc_preprocessor_res"] = self["doc_preprocessor_res"].json["res"]
         data["layout_det_res"] = self["layout_det_res"].json["res"]
-        if model_settings["use_general_ocr"] or model_settings["use_table_recognition"]:
-            data["overall_ocr_res"] = self["overall_ocr_res"].json["res"]
+        data["overall_ocr_res"] = self["overall_ocr_res"].json["res"]
         if model_settings["use_table_recognition"] and len(self["table_res_list"]) > 0:
             data["table_res_list"] = []
             for sno in range(len(self["table_res_list"])):
@@ -240,228 +389,357 @@ class LayoutParsingResultV2(BaseCVResult, HtmlMixin, XlsxMixin, MarkdownMixin):
                 res_xlsx_dict[key] = table_res.xlsx["pred"]
         return res_xlsx_dict
-    def _to_markdown(self) -> dict:
+    def _to_markdown(self, pretty=True) -> dict:
         """
         Save the parsing result to a Markdown file.
+        Args:
+            pretty (Optional[bool]): whether to pretty markdown by HTML, default by True.
         Returns:
             Dict
         """
+        original_image_width = self["doc_preprocessor_res"]["output_img"].shape[1]
-        def _format_data(obj):
-            def format_title(title):
-                """
-                Normalize chapter title.
-                Add the '#' to indicate the level of the title.
-                If numbering exists, ensure there's exactly one space between it and the title content.
-                If numbering does not exist, return the original title unchanged.
-                :param title: Original chapter title string.
-                :return: Normalized chapter title string.
-                """
-                match = self.title_pattern.match(title)
-                if match:
-                    numbering = match.group(1).strip()
-                    title_content = match.group(3).lstrip()
-                    # Return numbering and title content separated by one space
-                    title = numbering + " " + title_content
-                title = title.rstrip(".")
-                level = (
-                    title.count(
-                        ".",
-                    )
-                    + 1
-                    if "." in title
-                    else 1
-                )
-                return f"#{'#' * level} {title}".replace("-\n", "").replace(
-                    "\n",
-                    " ",
-                )
-            def format_centered_text(key):
-                return (
-                    f'<div style="text-align: center;">{block[key]}</div>'.replace(
-                        "-\n",
-                        "",
-                    ).replace("\n", " ")
-                    + "\n"
+        if pretty:
+            format_text_func = lambda block: format_centered_by_html(
+                format_text_plain_func(block)
+            )
+            format_image_func = lambda block: format_centered_by_html(
+                format_image_scaled_by_html_func(
+                    block,
+                    original_image_width=original_image_width,
                 )
+            )
+        else:
+            format_text_func = lambda block: block.content
+            format_image_func = format_image_plain_func
-            def format_image(label):
-                img_tags = []
-                image_path = "".join(block[label].keys())
-                img_tags.append(
-                    '<div style="text-align: center;"><img src="{}" alt="Image" /></div>'.format(
-                        image_path.replace("-\n", "").replace("\n", " "),
-                    ),
-                )
-                return "\n".join(img_tags)
-            def format_first_line(templates, format_func, spliter):
-                lines = block["block_content"].split(spliter)
-                for idx in range(len(lines)):
-                    line = lines[idx]
-                    if line.strip() == "":
-                        continue
-                    if line.lower() in templates:
-                        lines[idx] = format_func(line)
-                    break
-                return spliter.join(lines)
-            def format_table():
-                return "\n" + block["block_content"]
-            def get_seg_flag(block, prev_block):
-                seg_start_flag = True
-                seg_end_flag = True
-                block_box = block["block_bbox"]
-                context_left_coordinate = block_box[0]
-                context_right_coordinate = block_box[2]
-                seg_start_coordinate = block.get("seg_start_coordinate")
-                seg_end_coordinate = block.get("seg_end_coordinate")
-                if prev_block is not None:
-                    prev_block_bbox = prev_block["block_bbox"]
-                    num_of_prev_lines = prev_block.get("num_of_lines")
-                    pre_block_seg_end_coordinate = prev_block.get("seg_end_coordinate")
-                    prev_end_space_small = (
-                        context_right_coordinate - pre_block_seg_end_coordinate < 10
-                    )
-                    prev_lines_more_than_one = num_of_prev_lines > 1
-                    overlap_blocks = context_left_coordinate < prev_block_bbox[2]
-                    # update context_left_coordinate and context_right_coordinate
-                    if overlap_blocks:
-                        context_left_coordinate = min(
-                            prev_block_bbox[0], context_left_coordinate
-                        )
-                        context_right_coordinate = max(
-                            prev_block_bbox[2], context_right_coordinate
-                        )
-                        prev_end_space_small = (
-                            prev_block_bbox[2] - pre_block_seg_end_coordinate < 10
-                        )
-                    current_start_space_small = (
-                        seg_start_coordinate - context_left_coordinate < 10
-                    )
+        if self["model_settings"].get("use_chart_recognition", False):
+            format_chart_func = format_chart2table_func
+        else:
+            format_chart_func = format_image_func
-                    if (
-                        prev_end_space_small
-                        and current_start_space_small
-                        and prev_lines_more_than_one
-                    ):
-                        seg_start_flag = False
-                else:
-                    if seg_start_coordinate - context_left_coordinate < 10:
-                        seg_start_flag = False
-                if context_right_coordinate - seg_end_coordinate < 10:
-                    seg_end_flag = False
-                return seg_start_flag, seg_end_flag
-            handlers = {
-                "paragraph_title": lambda: format_title(block["block_content"]),
-                "doc_title": lambda: f"# {block['block_content']}".replace(
-                    "-\n",
-                    "",
-                ).replace("\n", " "),
-                "table_title": lambda: format_centered_text("block_content"),
-                "figure_title": lambda: format_centered_text("block_content"),
-                "chart_title": lambda: format_centered_text("block_content"),
-                "text": lambda: block["block_content"]
-                .replace("-\n", " ")
-                .replace("\n", " "),
-                "abstract": lambda: format_first_line(
-                    ["摘要", "abstract"], lambda l: f"## {l}\n", " "
-                ),
-                "content": lambda: block["block_content"]
-                .replace("-\n", "  \n")
-                .replace("\n", "  \n"),
-                "image": lambda: format_image("block_image"),
-                "chart": lambda: format_image("block_image"),
-                "formula": lambda: f"$${block['block_content']}$$",
-                "table": format_table,
-                "reference": lambda: format_first_line(
-                    ["参考文献", "references"], lambda l: f"## {l}", "\n"
-                ),
-                "algorithm": lambda: block["block_content"].strip("\n"),
-                "seal": lambda: f"Words of Seals:\n{block['block_content']}",
-            }
-            parsing_res_list = obj["parsing_res_list"]
-            markdown_content = ""
-            last_label = None
-            seg_start_flag = None
-            seg_end_flag = None
-            prev_block = None
-            page_first_element_seg_start_flag = None
-            page_last_element_seg_end_flag = None
-            parsing_res_list = sorted(
-                parsing_res_list,
-                key=lambda x: x.get("sub_index", 999),
+        if self["model_settings"].get("use_seal_recognition", False):
+            format_seal_func = lambda block: "\n".join(
+                [format_image_func(block), format_text_func(block)]
             )
-            for block in parsing_res_list:
-                seg_start_flag, seg_end_flag = get_seg_flag(block, prev_block)
-                label = block.get("block_label")
-                page_first_element_seg_start_flag = (
-                    seg_start_flag
-                    if (page_first_element_seg_start_flag is None)
-                    else page_first_element_seg_start_flag
+        else:
+            format_seal_func = format_image_func
+        if self["model_settings"].get("use_table_recognition", False):
+            if pretty:
+                format_table_func = lambda block: "\n" + format_text_func(
+                    block
+                ).replace("<table>", '<table border="1">')
+            else:
+                format_table_func = lambda block: simplify_table_func(
+                    "\n" + block.content
                 )
-                handler = handlers.get(label)
-                if handler:
-                    prev_block = block
-                    if label == last_label == "text" and seg_start_flag == False:
-                        last_char_of_markdown = (
-                            markdown_content[-1] if markdown_content else ""
-                        )
-                        first_char_of_handler = handler()[0] if handler() else ""
-                        last_is_chinese_char = (
-                            re.match(r"[\u4e00-\u9fff]", last_char_of_markdown)
-                            if last_char_of_markdown
-                            else False
-                        )
-                        first_is_chinese_char = (
-                            re.match(r"[\u4e00-\u9fff]", first_char_of_handler)
-                            if first_char_of_handler
-                            else False
-                        )
-                        if not (last_is_chinese_char or first_is_chinese_char):
-                            markdown_content += " " + handler()
-                        else:
-                            markdown_content += handler()
-                    else:
-                        markdown_content += (
-                            "\n\n" + handler() if markdown_content else handler()
-                        )
-                    last_label = label
-            page_last_element_seg_end_flag = seg_end_flag
-            return markdown_content, (
-                page_first_element_seg_start_flag,
-                page_last_element_seg_end_flag,
+        else:
+            format_table_func = format_image_func
+        if self["model_settings"].get("use_formula_recognition", False):
+            format_formula_func = lambda block: f"$${block.content}$$"
+        else:
+            format_formula_func = format_image_func
+        handle_funcs_dict = {
+            "paragraph_title": format_title_func,
+            "abstract_title": format_title_func,
+            "reference_title": format_title_func,
+            "content_title": format_title_func,
+            "doc_title": lambda block: f"# {block.content}".replace(
+                "-\n",
+                "",
+            ).replace("\n", " "),
+            "table_title": format_text_func,
+            "figure_title": format_text_func,
+            "chart_title": format_text_func,
+            "text": lambda block: block.content.replace("\n\n", "\n").replace(
+                "\n", "\n\n"
+            ),
+            "abstract": partial(
+                format_first_line_func,
+                templates=["摘要", "abstract"],
+                format_func=lambda l: f"## {l}\n",
+                spliter=" ",
+            ),
+            "content": lambda block: block.content.replace("-\n", "  \n").replace(
+                "\n", "  \n"
+            ),
+            "image": format_image_func,
+            "chart": format_chart_func,
+            "formula": format_formula_func,
+            "table": format_table_func,
+            "reference": partial(
+                format_first_line_func,
+                templates=["参考文献", "references"],
+                format_func=lambda l: f"## {l}",
+                spliter="\n",
+            ),
+            "algorithm": lambda block: block.content.strip("\n"),
+            "seal": format_seal_func,
+        }
+        markdown_content = ""
+        last_label = None
+        seg_start_flag = None
+        seg_end_flag = None
+        prev_block = None
+        page_first_element_seg_start_flag = None
+        page_last_element_seg_end_flag = None
+        markdown_info = {}
+        markdown_info["markdown_images"] = {}
+        for block in self["parsing_res_list"]:
+            seg_start_flag, seg_end_flag = get_seg_flag(block, prev_block)
+            label = block.label
+            if block.image is not None:
+                markdown_info["markdown_images"][block.image["path"]] = block.image[
+                    "img"
+                ]
+            page_first_element_seg_start_flag = (
+                seg_start_flag
+                if (page_first_element_seg_start_flag is None)
+                else page_first_element_seg_start_flag
             )
-        markdown_info = dict()
-        markdown_info["markdown_texts"], (
-            page_first_element_seg_start_flag,
-            page_last_element_seg_end_flag,
-        ) = _format_data(self)
+            handle_func = handle_funcs_dict.get(label, None)
+            if handle_func:
+                prev_block = block
+                if label == last_label == "text" and seg_start_flag == False:
+                    markdown_content += handle_func(block)
+                else:
+                    markdown_content += (
+                        "\n\n" + handle_func(block)
+                        if markdown_content
+                        else handle_func(block)
+                    )
+                last_label = label
+        page_last_element_seg_end_flag = seg_end_flag
+        markdown_info["markdown_texts"] = markdown_content
         markdown_info["page_continuation_flags"] = (
             page_first_element_seg_start_flag,
             page_last_element_seg_end_flag,
         )
-        markdown_info["markdown_images"] = {}
         for img in self["imgs_in_doc"]:
             markdown_info["markdown_images"][img["path"]] = img["img"]
         return markdown_info
+class LayoutParsingBlock:
+    def __init__(self, label, bbox, content="") -> None:
+        self.label = label
+        self.order_label = None
+        self.bbox = list(map(int, bbox))
+        self.content = content
+        self.seg_start_coordinate = float("inf")
+        self.seg_end_coordinate = float("-inf")
+        self.width = bbox[2] - bbox[0]
+        self.height = bbox[3] - bbox[1]
+        self.area = self.width * self.height
+        self.num_of_lines = 1
+        self.image = None
+        self.index = None
+        self.order_index = None
+        self.text_line_width = 1
+        self.text_line_height = 1
+        self.direction = self.get_bbox_direction()
+        self.child_blocks = []
+        self.update_direction_info()
+    def __str__(self) -> str:
+        return f"{self.__dict__}"
+    def __repr__(self) -> str:
+        _str = f"\n\n#################\nindex:\t{self.index}\nlabel:\t{self.label}\nregion_label:\t{self.order_label}\nbbox:\t{self.bbox}\ncontent:\t{self.content}\n#################"
+        return _str
+    def to_dict(self) -> dict:
+        return self.__dict__
+    def update_direction_info(self) -> None:
+        if self.direction == "horizontal":
+            self.secondary_direction = "vertical"
+            self.short_side_length = self.height
+            self.long_side_length = self.width
+            self.start_coordinate = self.bbox[0]
+            self.end_coordinate = self.bbox[2]
+            self.secondary_direction_start_coordinate = self.bbox[1]
+            self.secondary_direction_end_coordinate = self.bbox[3]
+        else:
+            self.secondary_direction = "horizontal"
+            self.short_side_length = self.width
+            self.long_side_length = self.height
+            self.start_coordinate = self.bbox[1]
+            self.end_coordinate = self.bbox[3]
+            self.secondary_direction_start_coordinate = self.bbox[0]
+            self.secondary_direction_end_coordinate = self.bbox[2]
+    def append_child_block(self, child_block: LayoutParsingBlock) -> None:
+        if not self.child_blocks:
+            self.ori_bbox = self.bbox.copy()
+        x1, y1, x2, y2 = self.bbox
+        x1_child, y1_child, x2_child, y2_child = child_block.bbox
+        union_bbox = (
+            min(x1, x1_child),
+            min(y1, y1_child),
+            max(x2, x2_child),
+            max(y2, y2_child),
+        )
+        self.bbox = union_bbox
+        self.update_direction_info()
+        child_blocks = [child_block]
+        if child_block.child_blocks:
+            child_blocks.extend(child_block.get_child_blocks())
+        self.child_blocks.extend(child_blocks)
+    def get_child_blocks(self) -> list:
+        self.bbox = self.ori_bbox
+        child_blocks = self.child_blocks.copy()
+        self.child_blocks = []
+        return child_blocks
+    def get_centroid(self) -> tuple:
+        x1, y1, x2, y2 = self.bbox
+        centroid = ((x1 + x2) / 2, (y1 + y2) / 2)
+        return centroid
+    def get_bbox_direction(self, direction_ratio: float = 1.0) -> bool:
+        """
+        Determine if a bounding box is horizontal or vertical.
+        Args:
+            bbox (List[float]): Bounding box [x_min, y_min, x_max, y_max].
+            direction_ratio (float): Ratio for determining direction. Default is 1.0.
+        Returns:
+            str: "horizontal" or "vertical".
+        """
+        return (
+            "horizontal" if self.width * direction_ratio >= self.height else "vertical"
+        )
+class LayoutParsingRegion:
+    def __init__(
+        self, bbox, blocks: List[LayoutParsingBlock] = [], image_shape=None
+    ) -> None:
+        self.bbox = bbox
+        self.block_map = {}
+        self.direction = "horizontal"
+        self.calculate_bbox_metrics(image_shape)
+        self.doc_title_block_idxes = []
+        self.paragraph_title_block_idxes = []
+        self.vision_block_idxes = []
+        self.unordered_block_idxes = []
+        self.vision_title_block_idxes = []
+        self.normal_text_block_idxes = []
+        self.header_block_idxes = []
+        self.footer_block_idxes = []
+        self.text_line_width = 20
+        self.text_line_height = 10
+        self.init_region_info_from_layout(blocks)
+        self.init_direction_info()
+    def init_region_info_from_layout(self, blocks: List[LayoutParsingBlock]):
+        horizontal_normal_text_block_num = 0
+        text_line_height_list = []
+        text_line_width_list = []
+        for idx, block in enumerate(blocks):
+            self.block_map[idx] = block
+            block.index = idx
+            if block.label in BLOCK_LABEL_MAP["header_labels"]:
+                self.header_block_idxes.append(idx)
+            elif block.label in BLOCK_LABEL_MAP["doc_title_labels"]:
+                self.doc_title_block_idxes.append(idx)
+            elif block.label in BLOCK_LABEL_MAP["paragraph_title_labels"]:
+                self.paragraph_title_block_idxes.append(idx)
+            elif block.label in BLOCK_LABEL_MAP["vision_labels"]:
+                self.vision_block_idxes.append(idx)
+            elif block.label in BLOCK_LABEL_MAP["vision_title_labels"]:
+                self.vision_title_block_idxes.append(idx)
+            elif block.label in BLOCK_LABEL_MAP["footer_labels"]:
+                self.footer_block_idxes.append(idx)
+            elif block.label in BLOCK_LABEL_MAP["unordered_labels"]:
+                self.unordered_block_idxes.append(idx)
+            else:
+                self.normal_text_block_idxes.append(idx)
+                text_line_height_list.append(block.text_line_height)
+                text_line_width_list.append(block.text_line_width)
+                if block.direction == "horizontal":
+                    horizontal_normal_text_block_num += 1
+        self.direction = (
+            "horizontal"
+            if horizontal_normal_text_block_num
+            >= len(self.normal_text_block_idxes) * 0.5
+            else "vertical"
+        )
+        self.text_line_width = (
+            np.mean(text_line_width_list) if text_line_width_list else 20
+        )
+        self.text_line_height = (
+            np.mean(text_line_height_list) if text_line_height_list else 10
+        )
+    def init_direction_info(self):
+        if self.direction == "horizontal":
+            self.direction_start_index = 0
+            self.direction_end_index = 2
+            self.secondary_direction_start_index = 1
+            self.secondary_direction_end_index = 3
+            self.secondary_direction = "vertical"
+        else:
+            self.direction_start_index = 1
+            self.direction_end_index = 3
+            self.secondary_direction_start_index = 0
+            self.secondary_direction_end_index = 2
+            self.secondary_direction = "horizontal"
+        self.direction_center_coordinate = (
+            self.bbox[self.direction_start_index] + self.bbox[self.direction_end_index]
+        ) / 2
+        self.secondary_direction_center_coordinate = (
+            self.bbox[self.secondary_direction_start_index]
+            + self.bbox[self.secondary_direction_end_index]
+        ) / 2
+    def calculate_bbox_metrics(self, image_shape):
+        x1, y1, x2, y2 = self.bbox
+        image_height, image_width = image_shape
+        width = x2 - x1
+        x_center, y_center = (x1 + x2) / 2, (y1 + y2) / 2
+        self.euclidean_distance = math.sqrt(((x1) ** 2 + (y1) ** 2))
+        self.center_euclidean_distance = math.sqrt(((x_center) ** 2 + (y_center) ** 2))
+        self.angle_rad = math.atan2(y_center, x_center)
+        self.weighted_distance = (
+            y2 + width + (x1 // (image_width // 10)) * (image_width // 10) * 1.5
+        )
+    def sort_normal_blocks(self, blocks):
+        if self.direction == "horizontal":
+            blocks.sort(
+                key=lambda x: (
+                    x.bbox[1] // self.text_line_height,
+                    x.bbox[0] // self.text_line_width,
+                    x.bbox[1] ** 2 + x.bbox[0] ** 2,
+                ),
+            )
+        else:
+            blocks.sort(
+                key=lambda x: (
+                    -x.bbox[0] // self.text_line_width,
+                    x.bbox[1] // self.text_line_height,
+                    -(x.bbox[2] ** 2 + x.bbox[1] ** 2),
+                ),
+            )
+    def sort(self):
+        from .xycut_enhanced import xycut_enhanced
+        return xycut_enhanced(self)

paddlex 3.0.0rc1__py3-none-any.whl → 3.0.1__py3-none-any.whl

paddlex 3.0.0rc1py3-none-any.whl → 3.0.1py3-none-any.whl