mineru 2.7.2__py3-none-any.whl → 2.7.3__py3-none-any.whl

mineru/backend/hybrid/hybrid_magic_model.py

@@ -424,7 +424,8 @@ def __tie_up_category_by_index(blocks, subject_block_type, object_block_type):
     # 调用通用方法
     return tie_up_category_by_index(
         get_subjects,
-        get_objects
+        get_objects,
+        object_block_type=object_block_type
     )
 
 

mineru/backend/vlm/vlm_magic_model.py

@@ -349,7 +349,8 @@ def __tie_up_category_by_index(blocks, subject_block_type, object_block_type):
     # 调用通用方法
     return tie_up_category_by_index(
         get_subjects,
-        get_objects
+        get_objects,
+        object_block_type=object_block_type
     )
 
 

mineru/utils/magic_model_utils.py

@@ -2,6 +2,8 @@
 包含两个MagicModel类中重复使用的方法和逻辑
 """
 from typing import List, Dict, Any, Callable
+
+from loguru import logger
 from mineru.utils.boxbase import bbox_distance, bbox_center_distance, is_in
 
 
@@ -172,11 +174,15 @@ def tie_up_category_by_index(
         get_subjects_func: Callable,
         get_objects_func: Callable,
         extract_subject_func: Callable = None,
-        extract_object_func: Callable = None
+        extract_object_func: Callable = None,
+        object_block_type: str = "object",
 ):
     """
     基于index的类别关联方法，用于将主体对象与客体对象进行关联
-    客体优先匹配给index最接近的主体，index差值相同时使用bbox中心点距离作为tiebreaker
+    客体优先匹配给index最接近的主体，匹配优先级为：
+    1. index差值（最高优先级）
+    2. bbox边缘距离（相邻边距离）
+    3. bbox中心点距离（最低优先级，作为最终tiebreaker）
 
     参数:
         get_subjects_func: 函数，提取主体对象
@@ -207,6 +213,29 @@ def tie_up_category_by_index(
             "sub_idx": i,
         }
 
+    # 提取所有客体的index集合，用于计算有效index差值
+    object_indices = set(obj["index"] for obj in objects)
+
+    def calc_effective_index_diff(obj_index: int, sub_index: int) -> int:
+        """
+        计算有效的index差值
+        有效差值 = 绝对差值 - 区间内其他客体的数量
+        即：如果obj_index和sub_index之间的差值是由其他客体造成的，则应该扣除这部分差值
+        """
+        if obj_index == sub_index:
+            return 0
+
+        start, end = min(obj_index, sub_index), max(obj_index, sub_index)
+        abs_diff = end - start
+
+        # 计算区间(start, end)内有多少个其他客体的index
+        other_objects_count = 0
+        for idx in range(start + 1, end):
+            if idx in object_indices:
+                other_objects_count += 1
+
+        return abs_diff - other_objects_count
+
     # 为每个客体找到最匹配的主体
     for obj in objects:
         if len(subjects) == 0:
@@ -217,10 +246,10 @@ def tie_up_category_by_index(
         min_index_diff = float("inf")
         best_subject_indices = []
 
-        # 找出index差值最小的所有主体
+        # 找出有效index差值最小的所有主体
         for i, subject in enumerate(subjects):
             sub_index = subject["index"]
-            index_diff = abs(obj_index - sub_index)
+            index_diff = calc_effective_index_diff(obj_index, sub_index)
 
             if index_diff < min_index_diff:
                 min_index_diff = index_diff
@@ -228,18 +257,37 @@ def tie_up_category_by_index(
             elif index_diff == min_index_diff:
                 best_subject_indices.append(i)
 
-        # 如果有多个主体的index差值相同，使用中心点距离作为tiebreaker
-        if len(best_subject_indices) > 1:
-            min_center_dist = float("inf")
+        if len(best_subject_indices) == 1:
             best_subject_idx = best_subject_indices[0]
-
-            for idx in best_subject_indices:
-                center_dist = bbox_center_distance(obj["bbox"], subjects[idx]["bbox"])
-                if center_dist < min_center_dist:
-                    min_center_dist = center_dist
-                    best_subject_idx = idx
+        # 如果有多个主体的index差值相同（最多两个），根据边缘距离进行筛选
+        elif len(best_subject_indices) == 2:
+            # 计算所有候选主体的边缘距离
+            edge_distances = [(idx, bbox_distance(obj["bbox"], subjects[idx]["bbox"])) for idx in best_subject_indices]
+            edge_dist_diff = abs(edge_distances[0][1] - edge_distances[1][1])
+
+            for idx, edge_dist in edge_distances:
+                logger.debug(f"Obj index: {obj_index}, Sub index: {subjects[idx]['index']}, Edge distance: {edge_dist}")
+
+            if edge_dist_diff > 2:
+                # 边缘距离差值大于2，匹配边缘距离更小的主体
+                best_subject_idx = min(edge_distances, key=lambda x: x[1])[0]
+                logger.debug(f"Obj index: {obj_index}, edge_dist_diff > 2, matching to subject with min edge distance, index: {subjects[best_subject_idx]['index']}")
+            elif object_block_type == "table_caption":
+                # 边缘距离差值<=2且为table_caption，匹配index更大的主体
+                best_subject_idx = max(best_subject_indices, key=lambda idx: subjects[idx]["index"])
+                logger.debug(f"Obj index: {obj_index}, edge_dist_diff <= 2 and table_caption, matching to later subject with index: {subjects[best_subject_idx]['index']}")
+            elif object_block_type.endswith("footnote"):
+                # 边缘距离差值<=2且为footnote，匹配index更小的主体
+                best_subject_idx = min(best_subject_indices, key=lambda idx: subjects[idx]["index"])
+                logger.debug(f"Obj index: {obj_index}, edge_dist_diff <= 2 and footnote, matching to earlier subject with index: {subjects[best_subject_idx]['index']}")
+            else:
+                # 边缘距离差值<=2 且不适用特殊匹配规则，使用中心点距离匹配
+                center_distances = [(idx, bbox_center_distance(obj["bbox"], subjects[idx]["bbox"])) for idx in best_subject_indices]
+                for idx, center_dist in center_distances:
+                    logger.debug(f"Obj index: {obj_index}, Sub index: {subjects[idx]['index']}, Center distance: {center_dist}")
+                best_subject_idx = min(center_distances, key=lambda x: x[1])[0]
         else:
-            best_subject_idx = best_subject_indices[0]
+            raise ValueError("More than two subjects have the same minimal index difference, which is unexpected.")
 
         # 将客体添加到最佳主体的obj_bboxes中
         result_dict[best_subject_idx]["obj_bboxes"].append(extract_object_func(obj))

mineru/version.py

@@ -1 +1 @@
-__version__ = "2.7.2"
+__version__ = "2.7.3"

{mineru-2.7.2.dist-info → mineru-2.7.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mineru
-Version: 2.7.2
+Version: 2.7.3
 Summary: A practical tool for converting PDF to Markdown
 License: AGPL-3.0
 Project-URL: homepage, https://mineru.net/

{mineru-2.7.2.dist-info → mineru-2.7.3.dist-info}/RECORD

@@ -1,10 +1,10 @@
 mineru/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
-mineru/version.py,sha256=H1WLrviWKvrPzDle8EWdCYYkzljxs0mtbXigYc-xaKA,22
+mineru/version.py,sha256=uf6cgtzZWaYn5QApMyykHXMzWM_oEqWLhYTsWSWu2_k,22
 mineru/backend/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
 mineru/backend/utils.py,sha256=GLJU3IznDmhE1_qNmkU1UOtsuskIHBezgsEVO6Uar-Y,698
 mineru/backend/hybrid/__init__.py,sha256=IFgr2C8NfSAj8q7JF7QOqMvCiJ6Fc8TIuU3Uh2DaFZU,51
 mineru/backend/hybrid/hybrid_analyze.py,sha256=Sckw6T-pvMv3V_nqZkBeW8kY4zNIBlWxqeS2vXqNqtY,20939
-mineru/backend/hybrid/hybrid_magic_model.py,sha256=39ByeZh54KBbPe77bzGCqZrZ5RNwNxGYttcoisgDOrc,24668
+mineru/backend/hybrid/hybrid_magic_model.py,sha256=_DvBq5WP_UZvmHfhZloxqv-MKoWWe_ye1kNLv6RA5rU,24713
 mineru/backend/hybrid/hybrid_model_output_to_middle_json.py,sha256=yE-c1eGa5LzPqLfKfvBON_SJRljqyz2B7LiglFcE7FQ,8468
 mineru/backend/pipeline/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
 mineru/backend/pipeline/batch_analyze.py,sha256=3UBs2WOwcI-mfGAlxZt437OqSOleXPLnpYbrD9h5D54,21303
@@ -19,7 +19,7 @@ mineru/backend/vlm/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5
 mineru/backend/vlm/model_output_to_middle_json.py,sha256=AqYX44gS9crUO_t7SuUatD71EVjow6pI6yA2Ik3gQ0s,5139
 mineru/backend/vlm/utils.py,sha256=1qma_KmDjRfOckcPbriGgRhS1XMk_johsyACfwcmDr4,3844
 mineru/backend/vlm/vlm_analyze.py,sha256=ttnQBUy1PEm9JZoF2G1_z-7gA3MgUUUBhz6OypCb4_g,14765
-mineru/backend/vlm/vlm_magic_model.py,sha256=mD-irxboo2DmMu4QF1wnvbti2xdNyBmNflbB4a-TmsU,21402
+mineru/backend/vlm/vlm_magic_model.py,sha256=RodoVwNJhzjyuRLn5Io5gFMIX1NxCuuLzCbUxGaKV80,21447
 mineru/backend/vlm/vlm_middle_json_mkcontent.py,sha256=w-Szbm4HitR7MY4pinSCZZdXtPSqmtlU9cjNh4IOQyg,29499
 mineru/cli/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
 mineru/cli/client.py,sha256=mPNfMEShVG-ithmlJQ5nGRIad2gCZgUjBGHN7zAmLhQ,6978
@@ -171,7 +171,7 @@ mineru/utils/guess_suffix_or_lang.py,sha256=aUC2wAJwa5LH0SHxwTbOEJqVVgvpdUCWFF6o
 mineru/utils/hash_utils.py,sha256=UPS_8NRBmVumdyOv16Lmv6Ly2xK8OVDJEe5gG6gKIFk,857
 mineru/utils/language.py,sha256=7RT3mxSa7jdpoC5ySd7ZddHA7TO7UsnmDOWiYZAxuyg,1433
 mineru/utils/llm_aided.py,sha256=9WUytvxenSAuaWR4sTQhVPQ5h8pY0wVOH1O2sj_6dLs,5149
-mineru/utils/magic_model_utils.py,sha256=I6vdN56aqhQBGOasoWHiJbjnXsBwUojw6xFjbWZSHaU,8656
+mineru/utils/magic_model_utils.py,sha256=8Hv-BDk9Ez4TUx6hrVJ_675yZZggPj6Uib81lSpm0ig,11683
 mineru/utils/model_utils.py,sha256=w-jSN7Ilh27FlMjPpKNO6MPbo_dT5Ln7zCQcXaREl_k,19605
 mineru/utils/models_download_utils.py,sha256=UfjvwhxO6BkJHa5JSpEVNZ71GoLMPMmJpym3THET2T4,2957
 mineru/utils/ocr_utils.py,sha256=lPIrwNUib5mrzUkponRYHuUCdjV2qvETNLSzOLyflrU,15990
@@ -185,9 +185,9 @@ mineru/utils/run_async.py,sha256=rPeP4BCZerR8VByRDhiYzfZiahLVqoZEBVAS54dAjNg,128
 mineru/utils/span_block_fix.py,sha256=0eVQjJCrT03woRt9hoh6Uu42Tp1dacfGTv2x3B9qq94,8797
 mineru/utils/span_pre_proc.py,sha256=nu6Bh5TWPKFzHuFfbEs0Asr04M4xOL5IONz_8GJHn44,13862
 mineru/utils/table_merge.py,sha256=LORxz0THemCqH746FMViqEuLzM088M4HgIkEuwDIfNU,21393
-mineru-2.7.2.dist-info/licenses/LICENSE.md,sha256=jVa0BUaKrRH4erV2P5AeJ24I2WRv9chIGxditreJ6e0,34524
-mineru-2.7.2.dist-info/METADATA,sha256=w3qS7X-Wjvqz8Ra5fp0QH-Wvq_RbZHGyaVOL8WIrerw,36621
-mineru-2.7.2.dist-info/WHEEL,sha256=qELbo2s1Yzl39ZmrAibXA2jjPLUYfnVhUNTlyF1rq0Y,92
-mineru-2.7.2.dist-info/entry_points.txt,sha256=JbtrCPhx1T32s7TONUsteKg-24ZwRT1HSiFtW5jypVw,376
-mineru-2.7.2.dist-info/top_level.txt,sha256=zuGQfZcbsHv4I4oKI9gaKPqEWBFm6xJroKuug2LnKP8,7
-mineru-2.7.2.dist-info/RECORD,,
+mineru-2.7.3.dist-info/licenses/LICENSE.md,sha256=jVa0BUaKrRH4erV2P5AeJ24I2WRv9chIGxditreJ6e0,34524
+mineru-2.7.3.dist-info/METADATA,sha256=XDUBoY78vVkmR2TFpXk_frncPD6D_Ev067KuoRUJR2U,36621
+mineru-2.7.3.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+mineru-2.7.3.dist-info/entry_points.txt,sha256=JbtrCPhx1T32s7TONUsteKg-24ZwRT1HSiFtW5jypVw,376
+mineru-2.7.3.dist-info/top_level.txt,sha256=zuGQfZcbsHv4I4oKI9gaKPqEWBFm6xJroKuug2LnKP8,7
+mineru-2.7.3.dist-info/RECORD,,

{mineru-2.7.2.dist-info → mineru-2.7.3.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (80.10.1)
+Generator: setuptools (80.10.2)
 Root-Is-Purelib: true
 Tag: py3-none-any