Merge branch 'master' of https://github.com/myhloli/Magic-PDF

demo/ocr_demo.py

@@ -90,9 +90,10 @@ def ocr_parse_core(book_name, ocr_pdf_path, ocr_pdf_model_info, start_page_id=0,
 
 
 if __name__ == '__main__':
-    # pdf_path = r"/home/cxu/workspace/Magic-PDF/ocr_demo/j.1540-627x.2006.00176.x.pdf"
-    # json_file_path = r"/home/cxu/workspace/Magic-PDF/ocr_demo/j.1540-627x.2006.00176.x.json"
+    pdf_path = r"/home/cxu/workspace/Magic-PDF/ocr_demo/j.1540-627x.2006.00176.x.pdf"
+    json_file_path = r"/home/cxu/workspace/Magic-PDF/ocr_demo/j.1540-627x.2006.00176.x.json"
     # ocr_local_parse(pdf_path, json_file_path)
-    # book_name = "数学新星网/edu_00001236"
-    # ocr_online_parse(book_name)
+    book_name = "科数网/edu_00011318"
+    ocr_online_parse(book_name)
+    
     pass

magic_pdf/dict2md/ocr_mkcontent.py

@@ -72,18 +72,26 @@ def ocr_mk_mm_markdown_with_para(pdf_info_dict: dict):
     markdown = []
     for _, page_info in pdf_info_dict.items():
         paras_of_layout = page_info.get("para_blocks")
-        page_markdown = ocr_mk_mm_markdown_with_para_core(paras_of_layout)
+        page_markdown = ocr_mk_mm_markdown_with_para_core(paras_of_layout, "mm")
         markdown.extend(page_markdown)
     return '\n\n'.join(markdown)
 
 
+def ocr_mk_nlp_markdown_with_para(pdf_info_dict: dict):
+    markdown = []
+    for _, page_info in pdf_info_dict.items():
+        paras_of_layout = page_info.get("para_blocks")
+        page_markdown = ocr_mk_mm_markdown_with_para_core(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: dict):
     markdown_with_para_and_pagination = []
     for page_no, page_info in pdf_info_dict.items():
         paras_of_layout = page_info.get("para_blocks")
         if not paras_of_layout:
             continue
-        page_markdown = ocr_mk_mm_markdown_with_para_core(paras_of_layout)
+        page_markdown = ocr_mk_mm_markdown_with_para_core(paras_of_layout, "mm")
         markdown_with_para_and_pagination.append({
             'page_no': page_no,
             'md_content': '\n\n'.join(page_markdown)
@@ -91,7 +99,7 @@ def ocr_mk_mm_markdown_with_para_and_pagination(pdf_info_dict: dict):
     return markdown_with_para_and_pagination
 
 
-def ocr_mk_mm_markdown_with_para_core(paras_of_layout):
+def ocr_mk_mm_markdown_with_para_core(paras_of_layout, mode):
     page_markdown = []
     for paras in paras_of_layout:
         for para in paras:
@@ -99,6 +107,7 @@ def ocr_mk_mm_markdown_with_para_core(paras_of_layout):
             for line in para:
                 for span in line['spans']:
                     span_type = span.get('type')
+                    content = ''
                     if span_type == ContentType.Text:
                         content = split_long_words(span['content'])
                         # content = span['content']
@@ -107,9 +116,16 @@ def ocr_mk_mm_markdown_with_para_core(paras_of_layout):
                     elif span_type == ContentType.InterlineEquation:
                         content = f"\n$$\n{span['content']}\n$$\n"
                     elif span_type in [ContentType.Image, ContentType.Table]:
-                        content = f"\n![]({join_path(s3_image_save_path, span['image_path'])})\n"
-                    para_text += content + ' '
-            page_markdown.append(para_text.strip() + '  ')
+                        if mode == 'mm':
+                            content = f"\n![]({join_path(s3_image_save_path, span['image_path'])})\n"
+                        elif mode == 'nlp':
+                            pass
+                    if content != '':
+                        para_text += content + ' '
+            if para_text.strip() == '':
+                continue
+            else:
+                page_markdown.append(para_text.strip() + '  ')
     return page_markdown
 
 

magic_pdf/pdf_parse_by_ocr.py

@@ -57,16 +57,16 @@ def construct_page_component(blocks, layout_bboxes, page_id, page_w, page_h, lay
 
 
 def parse_pdf_by_ocr(
-    pdf_path,
-    s3_pdf_profile,
-    pdf_model_output,
-    save_path,
-    book_name,
-    pdf_model_profile=None,
-    image_s3_config=None,
-    start_page_id=0,
-    end_page_id=None,
-    debug_mode=False,
+        pdf_path,
+        s3_pdf_profile,
+        pdf_model_output,
+        save_path,
+        book_name,
+        pdf_model_profile=None,
+        image_s3_config=None,
+        start_page_id=0,
+        end_page_id=None,
+        debug_mode=False,
 ):
     pdf_bytes = read_file(pdf_path, s3_pdf_profile)
     save_tmp_path = os.path.join(os.path.dirname(__file__), "../..", "tmp", "unittest")
@@ -95,7 +95,6 @@ def parse_pdf_by_ocr(
 
     start_time = time.time()
 
-
     end_page_id = end_page_id if end_page_id else len(pdf_docs) - 1
     for page_id in range(start_page_id, end_page_id + 1):
 
@@ -125,13 +124,6 @@ def parse_pdf_by_ocr(
             page_id, page, ocr_page_info, md_bookname_save_path, debug_mode=debug_mode
         )
 
-        # 构建需要remove的bbox列表
-        # need_remove_spans_bboxes = []
-        # need_remove_spans_bboxes.extend(page_no_bboxes)
-        # need_remove_spans_bboxes.extend(header_bboxes)
-        # need_remove_spans_bboxes.extend(footer_bboxes)
-        # need_remove_spans_bboxes.extend(footnote_bboxes)
-
         # 构建需要remove的bbox字典
         need_remove_spans_bboxes_dict = {
             DropTag.PAGE_NUMBER: page_no_bboxes,
@@ -199,50 +191,48 @@ def parse_pdf_by_ocr(
             else:
                 continue
 
-
-
-
-        # 删除重叠spans中较小的那些
+        '''删除重叠spans中较小的那些'''
         spans, dropped_spans_by_span_overlap = remove_overlaps_min_spans(spans)
 
-        # 删除remove_span_block_bboxes中的bbox
-        # spans = remove_spans_by_bboxes(spans, need_remove_spans_bboxes)
-        # 按qa要求，增加drop相关数据
+        '''
+        删除remove_span_block_bboxes中的bbox
+        并增加drop相关数据
+        '''
         spans, dropped_spans_by_removed_bboxes = remove_spans_by_bboxes_dict(spans, need_remove_spans_bboxes_dict)
 
-        # 对image和table截图
+        '''对image和table截图'''
         spans = cut_image_and_table(spans, page, page_id, book_name, save_path, img_s3_client)
 
-        # 行内公式调整, 高度调整至与同行文字高度一致(优先左侧, 其次右侧)
+        '''行内公式调整, 高度调整至与同行文字高度一致(优先左侧, 其次右侧)'''
         displayed_list = []
         text_inline_lines = []
         modify_y_axis(spans, displayed_list, text_inline_lines)
-        # 模型识别错误的行间公式, type类型转换成行内公式
+
+        '''模型识别错误的行间公式, type类型转换成行内公式'''
         spans = modify_inline_equation(spans, displayed_list, text_inline_lines)
 
-        # bbox去除粘连
+        '''bbox去除粘连'''
         spans = remove_overlap_between_bbox(spans)
 
-        # 对tpye=["interline_equation", "image", "table"]进行额外处理,如果左边有字的话,将该span的bbox中y0调整至不高于文字的y0
+        '''
+        对tpye=["interline_equation", "image", "table"]进行额外处理,
+        如果左边有字的话,将该span的bbox中y0调整至不高于文字的y0
+        '''
         spans = adjust_bbox_for_standalone_block(spans)
 
-
-        # 从ocr_page_info中解析layout信息(按自然阅读方向排序,并修复重叠和交错的bad case)
+        '''从ocr_page_info中解析layout信息(按自然阅读方向排序,并修复重叠和交错的bad case)'''
         layout_bboxes, layout_tree = layout_detect(ocr_page_info['subfield_dets'], page, ocr_page_info)
 
-        # 将spans合并成line(在layout内,从上到下,从左到右)
+        '''将spans合并成line(在layout内,从上到下,从左到右)'''
         lines, dropped_spans_by_layout = merge_spans_to_line_by_layout(spans, layout_bboxes)
 
-        # 将lines合并成block
+        '''将lines合并成block'''
         blocks = merge_lines_to_block(lines)
 
-        # 根据block合并段落
-        #para_blocks = para_split(blocks, layout_bboxes)
-        
-        # 获取QA需要外置的list
+        '''获取QA需要外置的list'''
         images, tables, interline_equations, inline_equations = get_qa_need_list(blocks)
 
-        # drop的span_list合并
+        '''drop的span_list合并'''
         dropped_spans = []
         dropped_spans.extend(dropped_spans_by_span_overlap)
         dropped_spans.extend(dropped_spans_by_removed_bboxes)
@@ -263,19 +253,18 @@ def parse_pdf_by_ocr(
             elif span['type'] in [ContentType.InlineEquation, ContentType.InterlineEquation]:
                 dropped_equation_block.append(span)
 
-
-
-        # 构造pdf_info_dict
+        '''构造pdf_info_dict'''
         page_info = construct_page_component(blocks, layout_bboxes, page_id, page_w, page_h, layout_tree,
                                              images, tables, interline_equations, inline_equations,
-                                             dropped_text_block, dropped_image_block, dropped_table_block, dropped_equation_block,
+                                             dropped_text_block, dropped_image_block, dropped_table_block,
+                                             dropped_equation_block,
                                              need_remove_spans_bboxes_dict)
         pdf_info_dict[f"page_{page_id}"] = page_info
 
     """分段"""
-    para_split(pdf_info_dict)
-    
-    # 在测试时,保存调试信息
+    para_split(pdf_info_dict, debug_mode=debug_mode)
+
+    '''在测试时,保存调试信息'''
     if debug_mode:
         params_file_save_path = join_path(
             save_tmp_path, "md", book_name, "preproc_out.json"

magic_pdf/pipeline.py

@@ -7,7 +7,7 @@ from magic_pdf.dict2md.ocr_mkcontent import (
     ocr_mk_nlp_markdown,
     ocr_mk_mm_markdown,
     ocr_mk_mm_standard_format,
-    ocr_mk_mm_markdown_with_para, ocr_mk_mm_markdown_with_para_and_pagination,
+    ocr_mk_mm_markdown_with_para, ocr_mk_mm_markdown_with_para_and_pagination, ocr_mk_nlp_markdown_with_para,
 )
 from magic_pdf.libs.commons import (
     read_file,
@@ -510,7 +510,8 @@ def ocr_pdf_intermediate_dict_to_markdown_with_para(jso: dict, debug_mode=False)
         pdf_intermediate_dict = jso["pdf_intermediate_dict"]
         # 将 pdf_intermediate_dict 解压
         pdf_intermediate_dict = JsonCompressor.decompress_json(pdf_intermediate_dict)
-        markdown_content = ocr_mk_mm_markdown_with_para(pdf_intermediate_dict)
+        # markdown_content = ocr_mk_mm_markdown_with_para(pdf_intermediate_dict)
+        markdown_content = ocr_mk_nlp_markdown_with_para(pdf_intermediate_dict)
         jso["content"] = markdown_content
         logger.info(
             f"book_name is:{get_data_source(jso)}/{jso['file_id']},markdown content length is {len(markdown_content)}",