lkw

demo/draw_bbox.py

@@ -1,7 +1,9 @@
-from magic_pdf.libs.commons import fitz  # PyMuPDF
+from pathlib import Path
+
+from magic_pdf.libs.commons import fitz, join_path  # PyMuPDF
 from magic_pdf.pdf_parse_by_ocr import parse_pdf_by_ocr
 import json
-
+import os
 
 
 
@@ -20,7 +22,19 @@ doc = fitz.open(pdf_path)  # Open the PDF
 data = [[[-2, 0, 603, 80, 24]], [[-3, 0, 602, 80, 24]]]
 ocr_json_file_path = r"D:\projects\Magic-PDF\ocr_demo\ocr_0.json"
 ocr_pdf_info = read_json_file(ocr_json_file_path)
-pdf_info_dict = parse_pdf_by_ocr(ocr_pdf_info)
+
+pth = Path(ocr_json_file_path)
+book_name = pth.name
+save_tmp_path = os.path.join(os.path.dirname(__file__), "../..", "tmp", "unittest")
+save_path = join_path(save_tmp_path, "md")
+
+pdf_info_dict = parse_pdf_by_ocr(
+            pdf_path,
+            None,
+            ocr_pdf_info,
+            save_path,
+            book_name,
+            debug_mode=True)
 data_list = []
 for page in pdf_info_dict.values():
     page_list = []

magic_pdf/pdf_parse_by_ocr.py

@@ -4,7 +4,6 @@ import time
 
 from loguru import logger
 
-from magic_pdf.libs.ocr_dict_merge import merge_spans_to_line, remove_overlaps_min_spans, modify_y_axis
 from magic_pdf.libs.commons import read_file, join_path, fitz, get_img_s3_client, get_delta_time, get_docx_model_output
 from magic_pdf.libs.coordinate_transform import get_scale_ratio
 from magic_pdf.libs.safe_filename import sanitize_filename
@@ -14,7 +13,7 @@ from magic_pdf.pre_proc.detect_header import parse_headers
 from magic_pdf.pre_proc.detect_page_number import parse_pageNos
 from magic_pdf.pre_proc.ocr_cut_image import cut_image_and_table
 from magic_pdf.pre_proc.ocr_detect_layout import layout_detect
-from magic_pdf.pre_proc.ocr_dict_merge import remove_overlaps_min_spans, merge_spans_to_line_by_layout
+from magic_pdf.pre_proc.ocr_dict_merge import remove_overlaps_min_spans, merge_spans_to_line_by_layout, modify_y_axis
 from magic_pdf.pre_proc.ocr_remove_spans import remove_spans_by_bboxes
 
 
@@ -150,7 +149,7 @@ def parse_pdf_by_ocr(
         spans = remove_overlaps_min_spans(spans)
 
         # 对tpye=["displayed_equation", "image", "table"]进行额外处理,如果左边有字的话,将该span的bbox中y0调整低于文字的y0
-        # spans = modify_y_axis(spans)
+        spans = modify_y_axis(spans)
 
         # 删除remove_span_block_bboxes中的bbox
         spans = remove_spans_by_bboxes(spans, need_remove_spans_bboxes)

magic_pdf/pre_proc/ocr_dict_merge.py

@@ -113,15 +113,19 @@ def modify_y_axis(spans: list):
     #用于给行间公式搜索
     text_inline_lines = []
     for span in spans[1:]:
+        if span.get("content","") == "78.":
+            print("debug")
         # 如果当前的span类型为"displayed_equation" 或者 当前行中已经有"displayed_equation"
         # image和table类型，同上
         if span['type'] in ["displayed_equation", "image", "table"] or any(
                 s['type'] in ["displayed_equation", "image", "table"] for s in current_line):
             #传入
-            if spans[0]["type"] in ["displayed_equation", "image", "table"]:
+            if span["type"] in ["displayed_equation", "image", "table"]:
                 displayed_list.append(span)
             # 则开始新行
             lines.append(current_line)
+            if len(current_line) > 1 or current_line[0]["type"] in ["text", "inline_equation"]:
+                text_inline_lines.append((current_line, (line_first_y0, line_first_y)))
             current_line = [span]
             line_first_y0 = span["bbox"][1]
             line_first_y = span["bbox"][3]
@@ -140,15 +144,14 @@ def modify_y_axis(spans: list):
             lines.append(current_line)
             text_inline_lines.append((current_line, (line_first_y0, line_first_y)))
             current_line = [span]
-            line_first_y0 = spans[0]["bbox"][1]
-            line_first_y = spans[0]["bbox"][3]
+            line_first_y0 = span["bbox"][1]
+            line_first_y = span["bbox"][3]
 
         # 添加最后一行
     if current_line:
         lines.append(current_line)
-        if len(current_line)>1 or current_line[0]["type"] in ["text", "inline_equation"]:
+        if len(current_line) > 1 or current_line[0]["type"] in ["text", "inline_equation"]:
             text_inline_lines.append((current_line, (line_first_y0, line_first_y)))
-
     for line in text_inline_lines:
         # 按照x0坐标排序
         current_line = line[0]
@@ -164,14 +167,17 @@ def modify_y_axis(spans: list):
     #错误行间公式转行内公式
     j = 0
     for i in range(len(displayed_list)):
+        if i == 8:
+            print("debug")
         span = displayed_list[i]
         span_y0, span_y = span["bbox"][1], span["bbox"][3]
+
         while j < len(text_inline_lines):
             text_line = text_inline_lines[j]
             y0, y1 = text_line[1]
-            if span_y0 < y0 and span_y > y0 or span_y0 < y1 and span_y > y1 or span_y0 < y0 and span_y > y1 and __is_overlaps_y_exceeds_threshold(span['bbox'], (0, y0, 0, y1)):
+            if (span_y0 < y0 and span_y > y0 or span_y0 < y1 and span_y > y1 or span_y0 < y0 and span_y > y1) and __is_overlaps_y_exceeds_threshold(span['bbox'], (0, y0, 0, y1)):
                 span["bbox"][1] = y0
-                span["bbox"][3] = y1
+                # span["bbox"][3] = y1
                 if span["type"] == "displayed_equation":
                     span["type"] = "inline_equation"
                 break