magic-pdf 0.8.1__py3-none-any.whl → 0.9.1__py3-none-any.whl

magic_pdf/dict2md/ocr_mkcontent.py

@@ -1,6 +1,5 @@
 import re
 
-import wordninja
 from loguru import logger
 
 from magic_pdf.libs.commons import join_path
@@ -8,6 +7,7 @@ from magic_pdf.libs.language import detect_lang
 from magic_pdf.libs.MakeContentConfig import DropMode, MakeMode
 from magic_pdf.libs.markdown_utils import ocr_escape_special_markdown_char
 from magic_pdf.libs.ocr_content_type import BlockType, ContentType
+from magic_pdf.para.para_split_v3 import ListLineTag
 
 
 def __is_hyphen_at_line_end(line):
@@ -24,37 +24,6 @@ def __is_hyphen_at_line_end(line):
     return bool(re.search(r'[A-Za-z]+-\s*$', line))
 
 
-def split_long_words(text):
-    segments = text.split(' ')
-    for i in range(len(segments)):
-        words = re.findall(r'\w+|[^\w]', segments[i], re.UNICODE)
-        for j in range(len(words)):
-            if len(words[j]) > 10:
-                words[j] = ' '.join(wordninja.split(words[j]))
-        segments[i] = ''.join(words)
-    return ' '.join(segments)
-
-
-def ocr_mk_mm_markdown_with_para(pdf_info_list: list, img_buket_path):
-    markdown = []
-    for page_info in pdf_info_list:
-        paras_of_layout = page_info.get('para_blocks')
-        page_markdown = ocr_mk_markdown_with_para_core_v2(
-            paras_of_layout, 'mm', img_buket_path)
-        markdown.extend(page_markdown)
-    return '\n\n'.join(markdown)
-
-
-def ocr_mk_nlp_markdown_with_para(pdf_info_dict: list):
-    markdown = []
-    for page_info in pdf_info_dict:
-        paras_of_layout = page_info.get('para_blocks')
-        page_markdown = ocr_mk_markdown_with_para_core_v2(
-            paras_of_layout, 'nlp')
-        markdown.extend(page_markdown)
-    return '\n\n'.join(markdown)
-
-
 def ocr_mk_mm_markdown_with_para_and_pagination(pdf_info_dict: list,
                                                 img_buket_path):
     markdown_with_para_and_pagination = []
@@ -67,61 +36,23 @@ def ocr_mk_mm_markdown_with_para_and_pagination(pdf_info_dict: list,
             paras_of_layout, 'mm', img_buket_path)
         markdown_with_para_and_pagination.append({
             'page_no':
-            page_no,
+                page_no,
             'md_content':
-            '\n\n'.join(page_markdown)
+                '\n\n'.join(page_markdown)
         })
         page_no += 1
     return markdown_with_para_and_pagination
 
 
-def ocr_mk_markdown_with_para_core(paras_of_layout, mode, img_buket_path=''):
-    page_markdown = []
-    for paras in paras_of_layout:
-        for para in paras:
-            para_text = ''
-            for line in para:
-                for span in line['spans']:
-                    span_type = span.get('type')
-                    content = ''
-                    language = ''
-                    if span_type == ContentType.Text:
-                        content = span['content']
-                        language = detect_lang(content)
-                        if (language == 'en'):  # 只对英文长词进行分词处理，中文分词会丢失文本
-                            content = ocr_escape_special_markdown_char(
-                                split_long_words(content))
-                        else:
-                            content = ocr_escape_special_markdown_char(content)
-                    elif span_type == ContentType.InlineEquation:
-                        content = f"${span['content']}$"
-                    elif span_type == ContentType.InterlineEquation:
-                        content = f"\n$$\n{span['content']}\n$$\n"
-                    elif span_type in [ContentType.Image, ContentType.Table]:
-                        if mode == 'mm':
-                            content = f"\n![]({join_path(img_buket_path, span['image_path'])})\n"
-                        elif mode == 'nlp':
-                            pass
-                    if content != '':
-                        if language == 'en':  # 英文语境下 content间需要空格分隔
-                            para_text += content + ' '
-                        else:  # 中文语境下，content间不需要空格分隔
-                            para_text += content
-            if para_text.strip() == '':
-                continue
-            else:
-                page_markdown.append(para_text.strip() + '  ')
-    return page_markdown
-
-
 def ocr_mk_markdown_with_para_core_v2(paras_of_layout,
                                       mode,
-                                      img_buket_path=''):
+                                      img_buket_path='',
+                                      ):
     page_markdown = []
     for para_block in paras_of_layout:
         para_text = ''
         para_type = para_block['type']
-        if para_type == BlockType.Text:
+        if para_type in [BlockType.Text, BlockType.List, BlockType.Index]:
             para_text = merge_para_with_text(para_block)
         elif para_type == BlockType.Title:
             para_text = f'# {merge_para_with_text(para_block)}'
@@ -136,20 +67,21 @@ def ocr_mk_markdown_with_para_core_v2(paras_of_layout,
                         for line in block['lines']:
                             for span in line['spans']:
                                 if span['type'] == ContentType.Image:
-                                    para_text += f"\n![]({join_path(img_buket_path, span['image_path'])})  \n"
+                                    if span.get('image_path', ''):
+                                        para_text += f"\n![]({join_path(img_buket_path, span['image_path'])})  \n"
                 for block in para_block['blocks']:  # 2nd.拼image_caption
                     if block['type'] == BlockType.ImageCaption:
-                        para_text += merge_para_with_text(block)
-                for block in para_block['blocks']:  # 2nd.拼image_caption
+                        para_text += merge_para_with_text(block) + '  \n'
+                for block in para_block['blocks']:  # 3rd.拼image_footnote
                     if block['type'] == BlockType.ImageFootnote:
-                        para_text += merge_para_with_text(block)
+                        para_text += merge_para_with_text(block) + '  \n'
         elif para_type == BlockType.Table:
             if mode == 'nlp':
                 continue
             elif mode == 'mm':
                 for block in para_block['blocks']:  # 1st.拼table_caption
                     if block['type'] == BlockType.TableCaption:
-                        para_text += merge_para_with_text(block)
+                        para_text += merge_para_with_text(block) + '  \n'
                 for block in para_block['blocks']:  # 2nd.拼table_body
                     if block['type'] == BlockType.TableBody:
                         for line in block['lines']:
@@ -160,11 +92,11 @@ def ocr_mk_markdown_with_para_core_v2(paras_of_layout,
                                         para_text += f"\n\n$\n {span['latex']}\n$\n\n"
                                     elif span.get('html', ''):
                                         para_text += f"\n\n{span['html']}\n\n"
-                                    else:
+                                    elif span.get('image_path', ''):
                                         para_text += f"\n![]({join_path(img_buket_path, span['image_path'])})  \n"
                 for block in para_block['blocks']:  # 3rd.拼table_footnote
                     if block['type'] == BlockType.TableFootnote:
-                        para_text += merge_para_with_text(block)
+                        para_text += merge_para_with_text(block) + '  \n'
 
         if para_text.strip() == '':
             continue
@@ -174,22 +106,36 @@ def ocr_mk_markdown_with_para_core_v2(paras_of_layout,
     return page_markdown
 
 
-def merge_para_with_text(para_block):
-
-    def detect_language(text):
-        en_pattern = r'[a-zA-Z]+'
-        en_matches = re.findall(en_pattern, text)
-        en_length = sum(len(match) for match in en_matches)
-        if len(text) > 0:
-            if en_length / len(text) >= 0.5:
-                return 'en'
-            else:
-                return 'unknown'
+def detect_language(text):
+    en_pattern = r'[a-zA-Z]+'
+    en_matches = re.findall(en_pattern, text)
+    en_length = sum(len(match) for match in en_matches)
+    if len(text) > 0:
+        if en_length / len(text) >= 0.5:
+            return 'en'
         else:
-            return 'empty'
+            return 'unknown'
+    else:
+        return 'empty'
+
+
+# 连写字符拆分
+def __replace_ligatures(text: str):
+    text = re.sub(r'fi', 'fi', text)  # 替换 fi 连写符
+    text = re.sub(r'fl', 'fl', text)  # 替换 fl 连写符
+    text = re.sub(r'ff', 'ff', text)  # 替换 ff 连写符
+    text = re.sub(r'ffi', 'ffi', text)  # 替换 ffi 连写符
+    text = re.sub(r'ffl', 'ffl', text)  # 替换 ffl 连写符
+    return text
 
+
+def merge_para_with_text(para_block):
     para_text = ''
-    for line in para_block['lines']:
+    for i, line in enumerate(para_block['lines']):
+
+        if i >= 1 and line.get(ListLineTag.IS_LIST_START_LINE, False):
+            para_text += '  \n'
+
         line_text = ''
         line_lang = ''
         for span in line['spans']:
@@ -199,208 +145,120 @@ def merge_para_with_text(para_block):
         if line_text != '':
             line_lang = detect_lang(line_text)
         for span in line['spans']:
+
             span_type = span['type']
             content = ''
             if span_type == ContentType.Text:
-                content = span['content']
-                # language = detect_lang(content)
-                language = detect_language(content)
-                if language == 'en':  # 只对英文长词进行分词处理，中文分词会丢失文本
-                    content = ocr_escape_special_markdown_char(
-                        split_long_words(content))
-                else:
-                    content = ocr_escape_special_markdown_char(content)
+                content = ocr_escape_special_markdown_char(span['content'])
             elif span_type == ContentType.InlineEquation:
-                content = f" ${span['content']}$ "
+                content = f"${span['content']}$"
             elif span_type == ContentType.InterlineEquation:
                 content = f"\n$$\n{span['content']}\n$$\n"
 
+            content = content.strip()
             if content != '':
                 langs = ['zh', 'ja', 'ko']
                 if line_lang in langs:  # 遇到一些一个字一个span的文档，这种单字语言判断不准，需要用整行文本判断
-                    para_text += content  # 中文/日语/韩文语境下，content间不需要空格分隔
-                elif line_lang == 'en':
-                    # 如果是前一行带有-连字符，那么末尾不应该加空格
-                    if __is_hyphen_at_line_end(content):
-                        para_text += content[:-1]
-                    else:
-                        para_text += content + ' '
+                    if span_type in [ContentType.Text, ContentType.InterlineEquation]:
+                        para_text += content  # 中文/日语/韩文语境下，content间不需要空格分隔
+                    elif span_type == ContentType.InlineEquation:
+                        para_text += f" {content} "
                 else:
-                    para_text += content + ' '  # 西方文本语境下 content间需要空格分隔
-    return para_text
-
+                    if span_type in [ContentType.Text, ContentType.InlineEquation]:
+                        # 如果是前一行带有-连字符，那么末尾不应该加空格
+                        if __is_hyphen_at_line_end(content):
+                            para_text += content[:-1]
+                        elif len(content) == 1 and content not in ['A', 'I', 'a', 'i']:
+                            para_text += content
+                        else:  # 西方文本语境下 content间需要空格分隔
+                            para_text += f"{content} "
+                    elif span_type == ContentType.InterlineEquation:
+                        para_text += content
+            else:
+                continue
+    # 连写字符拆分
+    para_text = __replace_ligatures(para_text)
 
-def para_to_standard_format(para, img_buket_path):
-    para_content = {}
-    if len(para) == 1:
-        para_content = line_to_standard_format(para[0], img_buket_path)
-    elif len(para) > 1:
-        para_text = ''
-        inline_equation_num = 0
-        for line in para:
-            for span in line['spans']:
-                language = ''
-                span_type = span.get('type')
-                content = ''
-                if span_type == ContentType.Text:
-                    content = span['content']
-                    language = detect_lang(content)
-                    if language == 'en':  # 只对英文长词进行分词处理，中文分词会丢失文本
-                        content = ocr_escape_special_markdown_char(
-                            split_long_words(content))
-                    else:
-                        content = ocr_escape_special_markdown_char(content)
-                elif span_type == ContentType.InlineEquation:
-                    content = f"${span['content']}$"
-                    inline_equation_num += 1
-                if language == 'en':  # 英文语境下 content间需要空格分隔
-                    para_text += content + ' '
-                else:  # 中文语境下，content间不需要空格分隔
-                    para_text += content
-        para_content = {
-            'type': 'text',
-            'text': para_text,
-            'inline_equation_num': inline_equation_num,
-        }
-    return para_content
+    return para_text
 
 
-def para_to_standard_format_v2(para_block, img_buket_path, page_idx):
+def para_to_standard_format_v2(para_block, img_buket_path, page_idx, drop_reason=None):
     para_type = para_block['type']
-    if para_type == BlockType.Text:
+    para_content = {}
+    if para_type in [BlockType.Text, BlockType.List, BlockType.Index]:
         para_content = {
             'type': 'text',
             'text': merge_para_with_text(para_block),
-            'page_idx': page_idx,
         }
     elif para_type == BlockType.Title:
         para_content = {
             'type': 'text',
             'text': merge_para_with_text(para_block),
             'text_level': 1,
-            'page_idx': page_idx,
         }
     elif para_type == BlockType.InterlineEquation:
         para_content = {
             'type': 'equation',
             'text': merge_para_with_text(para_block),
             'text_format': 'latex',
-            'page_idx': page_idx,
         }
     elif para_type == BlockType.Image:
-        para_content = {'type': 'image', 'page_idx': page_idx}
+        para_content = {'type': 'image', 'img_path': '', 'img_caption': [], 'img_footnote': []}
         for block in para_block['blocks']:
             if block['type'] == BlockType.ImageBody:
-                para_content['img_path'] = join_path(
-                    img_buket_path,
-                    block['lines'][0]['spans'][0]['image_path'])
+                for line in block['lines']:
+                    for span in line['spans']:
+                        if span['type'] == ContentType.Image:
+                            if span.get('image_path', ''):
+                                para_content['img_path'] = join_path(img_buket_path, span['image_path'])
             if block['type'] == BlockType.ImageCaption:
-                para_content['img_caption'] = merge_para_with_text(block)
+                para_content['img_caption'].append(merge_para_with_text(block))
             if block['type'] == BlockType.ImageFootnote:
-                para_content['img_footnote'] = merge_para_with_text(block)
+                para_content['img_footnote'].append(merge_para_with_text(block))
     elif para_type == BlockType.Table:
-        para_content = {'type': 'table', 'page_idx': page_idx}
+        para_content = {'type': 'table', 'img_path': '', 'table_caption': [], 'table_footnote': []}
         for block in para_block['blocks']:
             if block['type'] == BlockType.TableBody:
-                if block["lines"][0]["spans"][0].get('latex', ''):
-                    para_content['table_body'] = f"\n\n$\n {block['lines'][0]['spans'][0]['latex']}\n$\n\n"
-                elif block["lines"][0]["spans"][0].get('html', ''):
-                    para_content['table_body'] = f"\n\n{block['lines'][0]['spans'][0]['html']}\n\n"
-                para_content['img_path'] = join_path(img_buket_path, block["lines"][0]["spans"][0]['image_path'])
+                for line in block['lines']:
+                    for span in line['spans']:
+                        if span['type'] == ContentType.Table:
+
+                            if span.get('latex', ''):
+                                para_content['table_body'] = f"\n\n$\n {span['latex']}\n$\n\n"
+                            elif span.get('html', ''):
+                                para_content['table_body'] = f"\n\n{span['html']}\n\n"
+
+                            if span.get('image_path', ''):
+                                para_content['img_path'] = join_path(img_buket_path, span['image_path'])
+
             if block['type'] == BlockType.TableCaption:
-                para_content['table_caption'] = merge_para_with_text(block)
+                para_content['table_caption'].append(merge_para_with_text(block))
             if block['type'] == BlockType.TableFootnote:
-                para_content['table_footnote'] = merge_para_with_text(block)
+                para_content['table_footnote'].append(merge_para_with_text(block))
 
-    return para_content
+    para_content['page_idx'] = page_idx
 
+    if drop_reason is not None:
+        para_content['drop_reason'] = drop_reason
 
-def make_standard_format_with_para(pdf_info_dict: list, img_buket_path: str):
-    content_list = []
-    for page_info in pdf_info_dict:
-        paras_of_layout = page_info.get('para_blocks')
-        if not paras_of_layout:
-            continue
-        for para_block in paras_of_layout:
-            para_content = para_to_standard_format_v2(para_block,
-                                                      img_buket_path)
-            content_list.append(para_content)
-    return content_list
-
-
-def line_to_standard_format(line, img_buket_path):
-    line_text = ''
-    inline_equation_num = 0
-    for span in line['spans']:
-        if not span.get('content'):
-            if not span.get('image_path'):
-                continue
-            else:
-                if span['type'] == ContentType.Image:
-                    content = {
-                        'type': 'image',
-                        'img_path': join_path(img_buket_path,
-                                              span['image_path']),
-                    }
-                    return content
-                elif span['type'] == ContentType.Table:
-                    content = {
-                        'type': 'table',
-                        'img_path': join_path(img_buket_path,
-                                              span['image_path']),
-                    }
-                    return content
-        else:
-            if span['type'] == ContentType.InterlineEquation:
-                interline_equation = span['content']
-                content = {
-                    'type': 'equation',
-                    'latex': f'$$\n{interline_equation}\n$$'
-                }
-                return content
-            elif span['type'] == ContentType.InlineEquation:
-                inline_equation = span['content']
-                line_text += f'${inline_equation}$'
-                inline_equation_num += 1
-            elif span['type'] == ContentType.Text:
-                text_content = ocr_escape_special_markdown_char(
-                    span['content'])  # 转义特殊符号
-                line_text += text_content
-    content = {
-        'type': 'text',
-        'text': line_text,
-        'inline_equation_num': inline_equation_num,
-    }
-    return content
-
-
-def ocr_mk_mm_standard_format(pdf_info_dict: list):
-    """content_list type         string
-    image/text/table/equation(行间的单独拿出来，行内的和text合并) latex        string
-    latex文本字段。 text         string      纯文本格式的文本数据。 md           string
-    markdown格式的文本数据。 img_path     string      s3://full/path/to/img.jpg."""
-    content_list = []
-    for page_info in pdf_info_dict:
-        blocks = page_info.get('preproc_blocks')
-        if not blocks:
-            continue
-        for block in blocks:
-            for line in block['lines']:
-                content = line_to_standard_format(line)
-                content_list.append(content)
-    return content_list
+    return para_content
 
 
 def union_make(pdf_info_dict: list,
                make_mode: str,
                drop_mode: str,
-               img_buket_path: str = ''):
+               img_buket_path: str = '',
+               ):
     output_content = []
     for page_info in pdf_info_dict:
+        drop_reason_flag = False
+        drop_reason = None
         if page_info.get('need_drop', False):
             drop_reason = page_info.get('drop_reason')
             if drop_mode == DropMode.NONE:
                 pass
+            elif drop_mode == DropMode.NONE_WITH_REASON:
+                drop_reason_flag = True
             elif drop_mode == DropMode.WHOLE_PDF:
                 raise Exception((f'drop_mode is {DropMode.WHOLE_PDF} ,'
                                  f'drop_reason is {drop_reason}'))
@@ -425,8 +283,12 @@ def union_make(pdf_info_dict: list,
             output_content.extend(page_markdown)
         elif make_mode == MakeMode.STANDARD_FORMAT:
             for para_block in paras_of_layout:
-                para_content = para_to_standard_format_v2(
-                    para_block, img_buket_path, page_idx)
+                if drop_reason_flag:
+                    para_content = para_to_standard_format_v2(
+                        para_block, img_buket_path, page_idx)
+                else:
+                    para_content = para_to_standard_format_v2(
+                        para_block, img_buket_path, page_idx)
                 output_content.append(para_content)
     if make_mode in [MakeMode.MM_MD, MakeMode.NLP_MD]:
         return '\n\n'.join(output_content)

magic_pdf/libs/Constants.py

@@ -10,18 +10,12 @@ block维度自定义字段
 # block中lines是否被删除
 LINES_DELETED = "lines_deleted"
 
-# struct eqtable
-STRUCT_EQTABLE = "struct_eqtable"
-
 # table recognition max time default value
 TABLE_MAX_TIME_VALUE = 400
 
 # pp_table_result_max_length
 TABLE_MAX_LEN = 480
 
-# pp table structure algorithm
-TABLE_MASTER = "TableMaster"
-
 # table master structure dict
 TABLE_MASTER_DICT = "table_master_structure_dict.txt"
 
@@ -29,12 +23,31 @@ TABLE_MASTER_DICT = "table_master_structure_dict.txt"
 TABLE_MASTER_DIR = "table_structure_tablemaster_infer/"
 
 # pp detect model dir
-DETECT_MODEL_DIR = "ch_PP-OCRv3_det_infer"
+DETECT_MODEL_DIR = "ch_PP-OCRv4_det_infer"
 
 # pp rec model dir
-REC_MODEL_DIR = "ch_PP-OCRv3_rec_infer"
+REC_MODEL_DIR = "ch_PP-OCRv4_rec_infer"
 
 # pp rec char dict path
 REC_CHAR_DICT = "ppocr_keys_v1.txt"
 
+# pp rec copy rec directory
+PP_REC_DIRECTORY = ".paddleocr/whl/rec/ch/ch_PP-OCRv4_rec_infer"
+
+# pp rec copy det directory
+PP_DET_DIRECTORY = ".paddleocr/whl/det/ch/ch_PP-OCRv4_det_infer"
+
+
+class MODEL_NAME:
+    # pp table structure algorithm
+    TABLE_MASTER = "tablemaster"
+    # struct eqtable
+    STRUCT_EQTABLE = "struct_eqtable"
+
+    DocLayout_YOLO = "doclayout_yolo"
+
+    LAYOUTLMv3 = "layoutlmv3"
+
+    YOLO_V8_MFD = "yolo_v8_mfd"
 
+    UniMerNet_v2_Small = "unimernet_small"

magic_pdf/libs/MakeContentConfig.py

@@ -8,3 +8,4 @@ class DropMode:
     WHOLE_PDF = "whole_pdf"
     SINGLE_PAGE = "single_page"
     NONE = "none"
+    NONE_WITH_REASON = "none_with_reason"

magic_pdf/libs/boxbase.py

@@ -445,3 +445,38 @@ def get_overlap_area(bbox1, bbox2):
 
     # The area of overlap area
     return (x_right - x_left) * (y_bottom - y_top)
+
+
+def calculate_vertical_projection_overlap_ratio(block1, block2):
+    """
+    Calculate the proportion of the x-axis covered by the vertical projection of two blocks.
+
+    Args:
+        block1 (tuple): Coordinates of the first block (x0, y0, x1, y1).
+        block2 (tuple): Coordinates of the second block (x0, y0, x1, y1).
+
+    Returns:
+        float: The proportion of the x-axis covered by the vertical projection of the two blocks.
+    """
+    x0_1, _, x1_1, _ = block1
+    x0_2, _, x1_2, _ = block2
+
+    # Calculate the intersection of the x-coordinates
+    x_left = max(x0_1, x0_2)
+    x_right = min(x1_1, x1_2)
+
+    if x_right < x_left:
+        return 0.0
+
+    # Length of the intersection
+    intersection_length = x_right - x_left
+
+    # Length of the x-axis projection of the first block
+    block1_length = x1_1 - x0_1
+
+    if block1_length == 0:
+        return 0.0
+
+    # Proportion of the x-axis covered by the intersection
+    # logger.info(f"intersection_length: {intersection_length}, block1_length: {block1_length}")
+    return intersection_length / block1_length

magic_pdf/libs/clean_memory.py

@@ -0,0 +1,10 @@
+# Copyright (c) Opendatalab. All rights reserved.
+import torch
+import gc
+
+
+def clean_memory():
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
+        torch.cuda.ipc_collect()
+    gc.collect()