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

magic_pdf/pipe/TXTPipe.py

@@ -11,19 +11,25 @@ from magic_pdf.user_api import parse_txt_pdf
 class TXTPipe(AbsPipe):
 
     def __init__(self, pdf_bytes: bytes, model_list: list, image_writer: AbsReaderWriter, is_debug: bool = False,
-                 start_page_id=0, end_page_id=None):
-        super().__init__(pdf_bytes, model_list, image_writer, is_debug, start_page_id, end_page_id)
+                 start_page_id=0, end_page_id=None, lang=None,
+                 layout_model=None, formula_enable=None, table_enable=None):
+        super().__init__(pdf_bytes, model_list, image_writer, is_debug, start_page_id, end_page_id, lang,
+                         layout_model, formula_enable, table_enable)
 
     def pipe_classify(self):
         pass
 
     def pipe_analyze(self):
         self.model_list = doc_analyze(self.pdf_bytes, ocr=False,
-                                      start_page_id=self.start_page_id, end_page_id=self.end_page_id)
+                                      start_page_id=self.start_page_id, end_page_id=self.end_page_id,
+                                      lang=self.lang, layout_model=self.layout_model,
+                                      formula_enable=self.formula_enable, table_enable=self.table_enable)
 
     def pipe_parse(self):
         self.pdf_mid_data = parse_txt_pdf(self.pdf_bytes, self.model_list, self.image_writer, is_debug=self.is_debug,
-                                          start_page_id=self.start_page_id, end_page_id=self.end_page_id)
+                                          start_page_id=self.start_page_id, end_page_id=self.end_page_id,
+                                          lang=self.lang, layout_model=self.layout_model,
+                                          formula_enable=self.formula_enable, table_enable=self.table_enable)
 
     def pipe_mk_uni_format(self, img_parent_path: str, drop_mode=DropMode.WHOLE_PDF):
         result = super().pipe_mk_uni_format(img_parent_path, drop_mode)

magic_pdf/pipe/UNIPipe.py

@@ -14,9 +14,11 @@ from magic_pdf.user_api import parse_union_pdf, parse_ocr_pdf
 class UNIPipe(AbsPipe):
 
     def __init__(self, pdf_bytes: bytes, jso_useful_key: dict, image_writer: AbsReaderWriter, is_debug: bool = False,
-                 start_page_id=0, end_page_id=None):
+                 start_page_id=0, end_page_id=None, lang=None,
+                 layout_model=None, formula_enable=None, table_enable=None):
         self.pdf_type = jso_useful_key["_pdf_type"]
-        super().__init__(pdf_bytes, jso_useful_key["model_list"], image_writer, is_debug, start_page_id, end_page_id)
+        super().__init__(pdf_bytes, jso_useful_key["model_list"], image_writer, is_debug, start_page_id, end_page_id,
+                         lang, layout_model, formula_enable, table_enable)
         if len(self.model_list) == 0:
             self.input_model_is_empty = True
         else:
@@ -28,22 +30,29 @@ class UNIPipe(AbsPipe):
     def pipe_analyze(self):
         if self.pdf_type == self.PIP_TXT:
             self.model_list = doc_analyze(self.pdf_bytes, ocr=False,
-                                          start_page_id=self.start_page_id, end_page_id=self.end_page_id)
+                                          start_page_id=self.start_page_id, end_page_id=self.end_page_id,
+                                          lang=self.lang, layout_model=self.layout_model,
+                                          formula_enable=self.formula_enable, table_enable=self.table_enable)
         elif self.pdf_type == self.PIP_OCR:
             self.model_list = doc_analyze(self.pdf_bytes, ocr=True,
-                                          start_page_id=self.start_page_id, end_page_id=self.end_page_id)
+                                          start_page_id=self.start_page_id, end_page_id=self.end_page_id,
+                                          lang=self.lang, layout_model=self.layout_model,
+                                          formula_enable=self.formula_enable, table_enable=self.table_enable)
 
     def pipe_parse(self):
         if self.pdf_type == self.PIP_TXT:
             self.pdf_mid_data = parse_union_pdf(self.pdf_bytes, self.model_list, self.image_writer,
                                                 is_debug=self.is_debug, input_model_is_empty=self.input_model_is_empty,
-                                                start_page_id=self.start_page_id, end_page_id=self.end_page_id)
+                                                start_page_id=self.start_page_id, end_page_id=self.end_page_id,
+                                                lang=self.lang, layout_model=self.layout_model,
+                                                formula_enable=self.formula_enable, table_enable=self.table_enable)
         elif self.pdf_type == self.PIP_OCR:
             self.pdf_mid_data = parse_ocr_pdf(self.pdf_bytes, self.model_list, self.image_writer,
                                               is_debug=self.is_debug,
-                                              start_page_id=self.start_page_id, end_page_id=self.end_page_id)
+                                              start_page_id=self.start_page_id, end_page_id=self.end_page_id,
+                                              lang=self.lang)
 
-    def pipe_mk_uni_format(self, img_parent_path: str, drop_mode=DropMode.WHOLE_PDF):
+    def pipe_mk_uni_format(self, img_parent_path: str, drop_mode=DropMode.NONE_WITH_REASON):
         result = super().pipe_mk_uni_format(img_parent_path, drop_mode)
         logger.info("uni_pipe mk content list finished")
         return result

magic_pdf/pre_proc/ocr_detect_all_bboxes.py

@@ -1,7 +1,7 @@
 from loguru import logger
 
 from magic_pdf.libs.boxbase import get_minbox_if_overlap_by_ratio, calculate_overlap_area_in_bbox1_area_ratio, \
-    calculate_iou
+    calculate_iou, calculate_vertical_projection_overlap_ratio
 from magic_pdf.libs.drop_tag import DropTag
 from magic_pdf.libs.ocr_content_type import BlockType
 from magic_pdf.pre_proc.remove_bbox_overlap import remove_overlap_between_bbox_for_block
@@ -60,6 +60,88 @@ def ocr_prepare_bboxes_for_layout_split(img_blocks, table_blocks, discarded_bloc
     return all_bboxes, all_discarded_blocks, drop_reasons
 
 
+def add_bboxes(blocks, block_type, bboxes):
+    for block in blocks:
+        x0, y0, x1, y1 = block['bbox']
+        if block_type in [
+            BlockType.ImageBody, BlockType.ImageCaption, BlockType.ImageFootnote,
+            BlockType.TableBody, BlockType.TableCaption, BlockType.TableFootnote
+        ]:
+            bboxes.append([x0, y0, x1, y1, None, None, None, block_type, None, None, None, None, block["score"], block["group_id"]])
+        else:
+            bboxes.append([x0, y0, x1, y1, None, None, None, block_type, None, None, None, None, block["score"]])
+
+
+def ocr_prepare_bboxes_for_layout_split_v2(
+        img_body_blocks, img_caption_blocks, img_footnote_blocks,
+        table_body_blocks, table_caption_blocks, table_footnote_blocks,
+        discarded_blocks, text_blocks, title_blocks, interline_equation_blocks, page_w, page_h
+):
+    all_bboxes = []
+
+    add_bboxes(img_body_blocks, BlockType.ImageBody, all_bboxes)
+    add_bboxes(img_caption_blocks, BlockType.ImageCaption, all_bboxes)
+    add_bboxes(img_footnote_blocks, BlockType.ImageFootnote, all_bboxes)
+    add_bboxes(table_body_blocks, BlockType.TableBody, all_bboxes)
+    add_bboxes(table_caption_blocks, BlockType.TableCaption, all_bboxes)
+    add_bboxes(table_footnote_blocks, BlockType.TableFootnote, all_bboxes)
+    add_bboxes(text_blocks, BlockType.Text, all_bboxes)
+    add_bboxes(title_blocks, BlockType.Title, all_bboxes)
+    add_bboxes(interline_equation_blocks, BlockType.InterlineEquation, all_bboxes)
+
+    '''block嵌套问题解决'''
+    '''文本框与标题框重叠，优先信任文本框'''
+    all_bboxes = fix_text_overlap_title_blocks(all_bboxes)
+    '''任何框体与舍弃框重叠，优先信任舍弃框'''
+    all_bboxes = remove_need_drop_blocks(all_bboxes, discarded_blocks)
+
+    # interline_equation 与title或text框冲突的情况，分两种情况处理
+    '''interline_equation框与文本类型框iou比较接近1的时候，信任行间公式框'''
+    all_bboxes = fix_interline_equation_overlap_text_blocks_with_hi_iou(all_bboxes)
+    '''interline_equation框被包含在文本类型框内，且interline_equation比文本区块小很多时信任文本框，这时需要舍弃公式框'''
+    # 通过后续大框套小框逻辑删除
+
+    '''discarded_blocks'''
+    all_discarded_blocks = []
+    add_bboxes(discarded_blocks, BlockType.Discarded, all_discarded_blocks)
+
+    '''footnote识别：宽度超过1/3页面宽度的，高度超过10的，处于页面下半50%区域的'''
+    footnote_blocks = []
+    for discarded in discarded_blocks:
+        x0, y0, x1, y1 = discarded['bbox']
+        if (x1 - x0) > (page_w / 3) and (y1 - y0) > 10 and y0 > (page_h / 2):
+            footnote_blocks.append([x0, y0, x1, y1])
+
+    '''移除在footnote下面的任何框'''
+    need_remove_blocks = find_blocks_under_footnote(all_bboxes, footnote_blocks)
+    if len(need_remove_blocks) > 0:
+        for block in need_remove_blocks:
+            all_bboxes.remove(block)
+            all_discarded_blocks.append(block)
+
+    '''经过以上处理后，还存在大框套小框的情况，则删除小框'''
+    all_bboxes = remove_overlaps_min_blocks(all_bboxes)
+    all_discarded_blocks = remove_overlaps_min_blocks(all_discarded_blocks)
+    '''将剩余的bbox做分离处理，防止后面分layout时出错'''
+    all_bboxes, drop_reasons = remove_overlap_between_bbox_for_block(all_bboxes)
+
+    return all_bboxes, all_discarded_blocks
+
+
+def find_blocks_under_footnote(all_bboxes, footnote_blocks):
+    need_remove_blocks = []
+    for block in all_bboxes:
+        block_x0, block_y0, block_x1, block_y1 = block[:4]
+        for footnote_bbox in footnote_blocks:
+            footnote_x0, footnote_y0, footnote_x1, footnote_y1 = footnote_bbox
+            # 如果footnote的纵向投影覆盖了block的纵向投影的80%且block的y0大于等于footnote的y1
+            if block_y0 >= footnote_y1 and calculate_vertical_projection_overlap_ratio((block_x0, block_y0, block_x1, block_y1), footnote_bbox) >= 0.8:
+                if block not in need_remove_blocks:
+                    need_remove_blocks.append(block)
+                    break
+    return need_remove_blocks
+
+
 def fix_interline_equation_overlap_text_blocks_with_hi_iou(all_bboxes):
     # 先提取所有text和interline block
     text_blocks = []

magic_pdf/pre_proc/ocr_dict_merge.py

@@ -49,8 +49,7 @@ def merge_spans_to_line(spans):
                 continue
 
             # 如果当前的span与当前行的最后一个span在y轴上重叠，则添加到当前行
-            if __is_overlaps_y_exceeds_threshold(span['bbox'],
-                                                 current_line[-1]['bbox']):
+            if __is_overlaps_y_exceeds_threshold(span['bbox'], current_line[-1]['bbox'], 0.5):
                 current_line.append(span)
             else:
                 # 否则，开始新行
@@ -154,6 +153,11 @@ def fill_spans_in_blocks(blocks, spans, radio):
             'type': block_type,
             'bbox': block_bbox,
         }
+        if block_type in [
+            BlockType.ImageBody, BlockType.ImageCaption, BlockType.ImageFootnote,
+            BlockType.TableBody, BlockType.TableCaption, BlockType.TableFootnote
+        ]:
+            block_dict["group_id"] = block[-1]
         block_spans = []
         for span in spans:
             span_bbox = span['bbox']
@@ -202,6 +206,27 @@ def fix_block_spans(block_with_spans, img_blocks, table_blocks):
     return fix_blocks
 
 
+def fix_block_spans_v2(block_with_spans):
+    """1、img_block和table_block因为包含caption和footnote的关系，存在block的嵌套关系
+    需要将caption和footnote的text_span放入相应img_block和table_block内的
+    caption_block和footnote_block中 2、同时需要删除block中的spans字段."""
+    fix_blocks = []
+    for block in block_with_spans:
+        block_type = block['type']
+
+        if block_type in [BlockType.Text, BlockType.Title,
+                          BlockType.ImageCaption, BlockType.ImageFootnote,
+                          BlockType.TableCaption, BlockType.TableFootnote
+                          ]:
+            block = fix_text_block(block)
+        elif block_type in [BlockType.InterlineEquation, BlockType.ImageBody, BlockType.TableBody]:
+            block = fix_interline_block(block)
+        else:
+            continue
+        fix_blocks.append(block)
+    return fix_blocks
+
+
 def fix_discarded_block(discarded_block_with_spans):
     fix_discarded_blocks = []
     for block in discarded_block_with_spans:

magic_pdf/resources/model_config/UniMERNet/demo.yaml

@@ -2,13 +2,13 @@ model:
   arch: unimernet
   model_type: unimernet
   model_config:
-    model_name: ./models
-    max_seq_len: 1024
-    length_aware: False
+    model_name: ./models/unimernet_base
+    max_seq_len: 1536
+
   load_pretrained: True
-  pretrained: ./models/pytorch_model.bin
+  pretrained: './models/unimernet_base/pytorch_model.pth'
   tokenizer_config:
-    path: ./models
+    path: ./models/unimernet_base
 
 datasets:
   formula_rec_eval:
@@ -18,7 +18,7 @@ datasets:
         image_size:
           - 192
           - 672
-   
+
 run:
   runner: runner_iter
   task: unimernet_train
@@ -43,4 +43,4 @@ run:
   distributed_type: ddp  # or fsdp when train llm
 
   generate_cfg:
-    temperature: 0.0
+    temperature: 0.0

magic_pdf/resources/model_config/model_configs.yaml

@@ -1,15 +1,7 @@
-config:
-  device: cpu
-  layout: True
-  formula: True
-  table_config:
-    model: TableMaster
-    is_table_recog_enable: False
-    max_time: 400
-
 weights:
-  layout: Layout/model_final.pth
-  mfd: MFD/weights.pt
-  mfr: MFR/UniMERNet
+  layoutlmv3: Layout/LayoutLMv3/model_final.pth
+  doclayout_yolo: Layout/YOLO/doclayout_yolo_ft.pt
+  yolo_v8_mfd: MFD/YOLO/yolo_v8_ft.pt
+  unimernet_small: MFR/unimernet_small
   struct_eqtable: TabRec/StructEqTable
-  TableMaster: TabRec/TableMaster
+  tablemaster: TabRec/TableMaster

magic_pdf/tools/cli.py

@@ -44,6 +44,18 @@ auto: automatically choose the best method for parsing pdf from ocr and txt.
 without method specified, auto will be used by default.""",
     default='auto',
 )
+@click.option(
+    '-l',
+    '--lang',
+    'lang',
+    type=str,
+    help="""
+    Input the languages in the pdf (if known) to improve OCR accuracy.  Optional.
+    You should input "Abbreviation" with language form url:
+    https://paddlepaddle.github.io/PaddleOCR/latest/en/ppocr/blog/multi_languages.html#5-support-languages-and-abbreviations
+    """,
+    default=None,
+)
 @click.option(
     '-d',
     '--debug',
@@ -68,7 +80,7 @@ without method specified, auto will be used by default.""",
     help='The ending page for PDF parsing, beginning from 0.',
     default=None,
 )
-def cli(path, output_dir, method, debug_able, start_page_id, end_page_id):
+def cli(path, output_dir, method, lang, debug_able, start_page_id, end_page_id):
     model_config.__use_inside_model__ = True
     model_config.__model_mode__ = 'full'
     os.makedirs(output_dir, exist_ok=True)
@@ -90,6 +102,7 @@ def cli(path, output_dir, method, debug_able, start_page_id, end_page_id):
                 debug_able,
                 start_page_id=start_page_id,
                 end_page_id=end_page_id,
+                lang=lang
             )
 
         except Exception as e:

magic_pdf/tools/common.py

@@ -6,8 +6,8 @@ import click
 from loguru import logger
 
 import magic_pdf.model as model_config
-from magic_pdf.libs.draw_bbox import (draw_layout_bbox, draw_span_bbox,
-                                      drow_model_bbox)
+from magic_pdf.libs.draw_bbox import (draw_layout_bbox, draw_line_sort_bbox,
+                                      draw_model_bbox, draw_span_bbox)
 from magic_pdf.libs.MakeContentConfig import DropMode, MakeMode
 from magic_pdf.pipe.OCRPipe import OCRPipe
 from magic_pdf.pipe.TXTPipe import TXTPipe
@@ -39,16 +39,21 @@ def do_parse(
     f_dump_middle_json=True,
     f_dump_model_json=True,
     f_dump_orig_pdf=True,
-    f_dump_content_list=False,
+    f_dump_content_list=True,
     f_make_md_mode=MakeMode.MM_MD,
     f_draw_model_bbox=False,
+    f_draw_line_sort_bbox=False,
     start_page_id=0,
     end_page_id=None,
+    lang=None,
+    layout_model=None,
+    formula_enable=None,
+    table_enable=None,
 ):
     if debug_able:
         logger.warning('debug mode is on')
-        f_dump_content_list = True
         f_draw_model_bbox = True
+        f_draw_line_sort_bbox = True
 
     orig_model_list = copy.deepcopy(model_list)
     local_image_dir, local_md_dir = prepare_env(output_dir, pdf_file_name,
@@ -61,13 +66,16 @@ def do_parse(
     if parse_method == 'auto':
         jso_useful_key = {'_pdf_type': '', 'model_list': model_list}
         pipe = UNIPipe(pdf_bytes, jso_useful_key, image_writer, is_debug=True,
-                       start_page_id=start_page_id, end_page_id=end_page_id)
+                       start_page_id=start_page_id, end_page_id=end_page_id, lang=lang,
+                       layout_model=layout_model, formula_enable=formula_enable, table_enable=table_enable)
     elif parse_method == 'txt':
         pipe = TXTPipe(pdf_bytes, model_list, image_writer, is_debug=True,
-                       start_page_id=start_page_id, end_page_id=end_page_id)
+                       start_page_id=start_page_id, end_page_id=end_page_id, lang=lang,
+                       layout_model=layout_model, formula_enable=formula_enable, table_enable=table_enable)
     elif parse_method == 'ocr':
         pipe = OCRPipe(pdf_bytes, model_list, image_writer, is_debug=True,
-                       start_page_id=start_page_id, end_page_id=end_page_id)
+                       start_page_id=start_page_id, end_page_id=end_page_id, lang=lang,
+                       layout_model=layout_model, formula_enable=formula_enable, table_enable=table_enable)
     else:
         logger.error('unknown parse method')
         exit(1)
@@ -89,7 +97,9 @@ def do_parse(
     if f_draw_span_bbox:
         draw_span_bbox(pdf_info, pdf_bytes, local_md_dir, pdf_file_name)
     if f_draw_model_bbox:
-        drow_model_bbox(copy.deepcopy(orig_model_list), pdf_bytes, local_md_dir, pdf_file_name)
+        draw_model_bbox(copy.deepcopy(orig_model_list), pdf_bytes, local_md_dir, pdf_file_name)
+    if f_draw_line_sort_bbox:
+        draw_line_sort_bbox(pdf_info, pdf_bytes, local_md_dir, pdf_file_name)
 
     md_content = pipe.pipe_mk_markdown(image_dir,
                                        drop_mode=DropMode.NONE,

magic_pdf/user_api.py

@@ -26,7 +26,7 @@ PARSE_TYPE_OCR = "ocr"
 
 
 def parse_txt_pdf(pdf_bytes: bytes, pdf_models: list, imageWriter: AbsReaderWriter, is_debug=False,
-                  start_page_id=0, end_page_id=None,
+                  start_page_id=0, end_page_id=None, lang=None,
                   *args, **kwargs):
     """
     解析文本类pdf
@@ -44,11 +44,14 @@ def parse_txt_pdf(pdf_bytes: bytes, pdf_models: list, imageWriter: AbsReaderWrit
 
     pdf_info_dict["_version_name"] = __version__
 
+    if lang is not None:
+        pdf_info_dict["_lang"] = lang
+
     return pdf_info_dict
 
 
 def parse_ocr_pdf(pdf_bytes: bytes, pdf_models: list, imageWriter: AbsReaderWriter, is_debug=False,
-                  start_page_id=0, end_page_id=None,
+                  start_page_id=0, end_page_id=None, lang=None,
                   *args, **kwargs):
     """
     解析ocr类pdf
@@ -66,12 +69,15 @@ def parse_ocr_pdf(pdf_bytes: bytes, pdf_models: list, imageWriter: AbsReaderWrit
 
     pdf_info_dict["_version_name"] = __version__
 
+    if lang is not None:
+        pdf_info_dict["_lang"] = lang
+
     return pdf_info_dict
 
 
 def parse_union_pdf(pdf_bytes: bytes, pdf_models: list, imageWriter: AbsReaderWriter, is_debug=False,
                     input_model_is_empty: bool = False,
-                    start_page_id=0, end_page_id=None,
+                    start_page_id=0, end_page_id=None, lang=None,
                     *args, **kwargs):
     """
     ocr和文本混合的pdf，全部解析出来
@@ -95,9 +101,19 @@ def parse_union_pdf(pdf_bytes: bytes, pdf_models: list, imageWriter: AbsReaderWr
     if pdf_info_dict is None or pdf_info_dict.get("_need_drop", False):
         logger.warning(f"parse_pdf_by_txt drop or error, switch to parse_pdf_by_ocr")
         if input_model_is_empty:
-            pdf_models = doc_analyze(pdf_bytes, ocr=True,
-                                     start_page_id=start_page_id,
-                                     end_page_id=end_page_id)
+            layout_model = kwargs.get("layout_model", None)
+            formula_enable = kwargs.get("formula_enable", None)
+            table_enable = kwargs.get("table_enable", None)
+            pdf_models = doc_analyze(
+                pdf_bytes,
+                ocr=True,
+                start_page_id=start_page_id,
+                end_page_id=end_page_id,
+                lang=lang,
+                layout_model=layout_model,
+                formula_enable=formula_enable,
+                table_enable=table_enable,
+            )
         pdf_info_dict = parse_pdf(parse_pdf_by_ocr)
         if pdf_info_dict is None:
             raise Exception("Both parse_pdf_by_txt and parse_pdf_by_ocr failed.")
@@ -108,4 +124,7 @@ def parse_union_pdf(pdf_bytes: bytes, pdf_models: list, imageWriter: AbsReaderWr
 
     pdf_info_dict["_version_name"] = __version__
 
+    if lang is not None:
+        pdf_info_dict["_lang"] = lang
+
     return pdf_info_dict

magic_pdf/utils/annotations.py

@@ -0,0 +1,11 @@
+
+from loguru import logger
+
+
+def ImportPIL(f):
+    try:
+        import PIL  # noqa: F401
+    except ImportError:
+        logger.error('Pillow not installed, please install by pip.')
+        exit(1)
+    return f

{magic_pdf-0.8.1.dist-info → magic_pdf-0.9.1.dist-info}/LICENSE.md

@@ -659,3 +659,4 @@ specific requirements.
 if any, to sign a "copyright disclaimer" for the program, if necessary.
 For more information on this, and how to apply and follow the GNU AGPL, see
 <https://www.gnu.org/licenses/>.
+