mineru 2.7.0__py3-none-any.whl → 2.7.2__py3-none-any.whl

mineru/backend/vlm/utils.py

@@ -18,6 +18,10 @@ def enable_custom_logits_processors() -> bool:
         compute_capability = f"{major}.{minor}"
     elif hasattr(torch, 'npu') and torch.npu.is_available():
         compute_capability = "8.0"
+    elif hasattr(torch, 'gcu') and torch.gcu.is_available():
+        compute_capability = "8.0"
+    elif hasattr(torch, 'musa') and torch.musa.is_available():
+        compute_capability = "8.0"
     else:
         logger.info("CUDA not available, disabling custom_logits_processors")
         return False

mineru/backend/vlm/vlm_analyze.py

@@ -1,6 +1,7 @@
 # Copyright (c) Opendatalab. All rights reserved.
 import os
 import time
+import json
 
 from loguru import logger
 
@@ -99,6 +100,30 @@ class ModelSingleton:
                         import vllm
                     except ImportError:
                         raise ImportError("Please install vllm to use the vllm-engine backend.")
+
+                    """
+                    # musa vllm v1 引擎特殊配置
+                    device = get_device()
+                    if device.startswith("musa"):
+                        import torch
+                        if torch.musa.is_available():
+                            compilation_config = {
+                                "cudagraph_capture_sizes": [1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 18, 20, 24, 28, 30],
+                                "simple_cuda_graph": True
+                            }
+                            block_size = 32
+                            kwargs["compilation_config"] = compilation_config
+                            kwargs["block_size"] = block_size
+                    """
+
+                    if "compilation_config" in kwargs:
+                        if isinstance(kwargs["compilation_config"], str):
+                            try:
+                                kwargs["compilation_config"] = json.loads(kwargs["compilation_config"])
+                            except json.JSONDecodeError:
+                                logger.warning(
+                                    f"Failed to parse compilation_config as JSON: {kwargs['compilation_config']}")
+                                del kwargs["compilation_config"]
                     if "gpu_memory_utilization" not in kwargs:
                         kwargs["gpu_memory_utilization"] = set_default_gpu_memory_utilization()
                     if "model" not in kwargs:
@@ -112,8 +137,38 @@ class ModelSingleton:
                     try:
                         from vllm.engine.arg_utils import AsyncEngineArgs
                         from vllm.v1.engine.async_llm import AsyncLLM
+                        from vllm.config import CompilationConfig
                     except ImportError:
                         raise ImportError("Please install vllm to use the vllm-async-engine backend.")
+
+                    """
+                    # musa vllm v1 引擎特殊配置
+                    device = get_device()
+                    if device.startswith("musa"):
+                        import torch
+                        if torch.musa.is_available():
+                            compilation_config = CompilationConfig(
+                                cudagraph_capture_sizes=[1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 18, 20, 24, 28, 30],
+                                simple_cuda_graph=True
+                            )
+                            block_size = 32
+                            kwargs["compilation_config"] = compilation_config
+                            kwargs["block_size"] = block_size
+                    """
+
+                    if "compilation_config" in kwargs:
+                        if isinstance(kwargs["compilation_config"], dict):
+                            # 如果是字典，转换为 CompilationConfig 对象
+                            kwargs["compilation_config"] = CompilationConfig(**kwargs["compilation_config"])
+                        elif isinstance(kwargs["compilation_config"], str):
+                            # 如果是 JSON 字符串，先解析再转换
+                            try:
+                                config_dict = json.loads(kwargs["compilation_config"])
+                                kwargs["compilation_config"] = CompilationConfig(**config_dict)
+                            except (json.JSONDecodeError, TypeError) as e:
+                                logger.warning(
+                                    f"Failed to parse compilation_config: {kwargs['compilation_config']}, error: {e}")
+                                del kwargs["compilation_config"]
                     if "gpu_memory_utilization" not in kwargs:
                         kwargs["gpu_memory_utilization"] = set_default_gpu_memory_utilization()
                     if "model" not in kwargs:

mineru/cli/common.py

@@ -17,8 +17,6 @@ from mineru.utils.pdf_image_tools import images_bytes_to_pdf_bytes
 from mineru.backend.vlm.vlm_middle_json_mkcontent import union_make as vlm_union_make
 from mineru.backend.vlm.vlm_analyze import doc_analyze as vlm_doc_analyze
 from mineru.backend.vlm.vlm_analyze import aio_doc_analyze as aio_vlm_doc_analyze
-from mineru.backend.hybrid.hybrid_analyze import doc_analyze as hybrid_doc_analyze
-from mineru.backend.hybrid.hybrid_analyze import aio_doc_analyze as aio_hybrid_doc_analyze
 from mineru.utils.pdf_page_id import get_end_page_id
 
 if os.getenv("MINERU_LMDEPLOY_DEVICE", "") == "maca":
@@ -326,6 +324,7 @@ def _process_hybrid(
         server_url=None,
         **kwargs,
 ):
+    from mineru.backend.hybrid.hybrid_analyze import doc_analyze as hybrid_doc_analyze
     """同步处理hybrid后端逻辑"""
     if not backend.endswith("client"):
         server_url = None
@@ -378,8 +377,8 @@ async def _async_process_hybrid(
         server_url=None,
         **kwargs,
 ):
+    from mineru.backend.hybrid.hybrid_analyze import aio_doc_analyze as aio_hybrid_doc_analyze
     """异步处理hybrid后端逻辑"""
-
     if not backend.endswith("client"):
         server_url = None
 

mineru/model/mfd/yolo_v8.py

@@ -1,5 +1,7 @@
 import os
 from typing import List, Union
+
+import torch
 from tqdm import tqdm
 from ultralytics import YOLO
 import numpy as np
@@ -18,8 +20,8 @@ class YOLOv8MFDModel:
         conf: float = 0.25,
         iou: float = 0.45,
     ):
-        self.model = YOLO(weight).to(device)
-        self.device = device
+        self.device = torch.device(device)
+        self.model = YOLO(weight).to(self.device)
         self.imgsz = imgsz
         self.conf = conf
         self.iou = iou

mineru/model/mfr/unimernet/Unimernet.py

@@ -23,12 +23,12 @@ class MathDataset(Dataset):
 class UnimernetModel(object):
     def __init__(self, weight_dir, _device_="cpu"):
         from .unimernet_hf import UnimernetModel
-        if _device_.startswith("mps") or _device_.startswith("npu"):
+        if _device_.startswith("mps") or _device_.startswith("npu") or _device_.startswith("musa"):
             self.model = UnimernetModel.from_pretrained(weight_dir, attn_implementation="eager")
         else:
             self.model = UnimernetModel.from_pretrained(weight_dir)
-        self.device = _device_
-        self.model.to(_device_)
+        self.device = torch.device(_device_)
+        self.model.to(self.device)
         if not _device_.startswith("cpu"):
             self.model = self.model.to(dtype=torch.float16)
         self.model.eval()

mineru/model/table/rec/unet_table/utils_table_line_rec.py

@@ -4,6 +4,8 @@ import cv2
 import numpy as np
 from scipy.spatial import distance as dist
 from skimage import measure
+from skimage import __version__ as skimage_version
+from packaging import version
 
 
 def transform_preds(coords, center, scale, output_size, rot=0):
@@ -295,7 +297,11 @@ def min_area_rect_box(
     """
     boxes = []
     for region in regions:
-        if region.bbox_area > H * W * 3 / 4:  # 过滤大的单元格
+        if version.parse(skimage_version) >= version.parse("0.26.0"):
+            region_bbox_area = region.area_bbox
+        else:
+            region_bbox_area = region.bbox_area
+        if region_bbox_area > H * W * 3 / 4:  # 过滤大的单元格
             continue
         rect = cv2.minAreaRect(region.coords[:, ::-1])
 

mineru/model/vlm/vllm_server.py

@@ -2,6 +2,7 @@ import os
 import sys
 
 from mineru.backend.vlm.utils import set_default_gpu_memory_utilization, enable_custom_logits_processors
+from mineru.utils.config_reader import get_device
 from mineru.utils.models_download_utils import auto_download_and_get_model_root_path
 
 from vllm.entrypoints.cli.main import main as vllm_main
@@ -13,6 +14,8 @@ def main():
     has_port_arg = False
     has_gpu_memory_utilization_arg = False
     has_logits_processors_arg = False
+    has_block_size_arg = False
+    has_compilation_config = False
     model_path = None
     model_arg_indices = []
 
@@ -24,6 +27,10 @@ def main():
             has_gpu_memory_utilization_arg = True
         if arg == "--logits-processors" or arg.startswith("--logits-processors="):
             has_logits_processors_arg = True
+        if arg == "--block-size" or arg.startswith("--block-size="):
+            has_block_size_arg = True
+        if arg == "--compilation-config" or arg.startswith("--compilation-config="):
+            has_compilation_config = True
         if arg == "--model":
             if i + 1 < len(args):
                 model_path = args[i + 1]
@@ -49,6 +56,17 @@ def main():
         model_path = auto_download_and_get_model_root_path("/", "vlm")
     if (not has_logits_processors_arg) and custom_logits_processors:
         args.extend(["--logits-processors", "mineru_vl_utils:MinerULogitsProcessor"])
+    """
+    # musa vllm v1 引擎特殊配置 
+    device = get_device()
+    if device.startswith("musa"):
+        import torch
+        if torch.musa.is_available():
+            if not has_block_size_arg:
+                args.extend(["--block-size", "32"])
+            if not has_compilation_config:
+                args.extend(["--compilation-config", '{"cudagraph_capture_sizes": [1,2,3,4,5,6,7,8,10,12,14,16,18,20,24,28,30], "simple_cuda_graph": true}'])
+    """
 
     # 重构参数，将模型路径作为位置参数
     sys.argv = [sys.argv[0]] + ["serve", model_path] + args

mineru/utils/block_sort.py

@@ -186,6 +186,18 @@ def model_init(model_name: str):
             bf_16_support = True
     elif device_name.startswith("mps"):
         bf_16_support = True
+    elif device_name.startswith("gcu"):
+        if hasattr(torch, 'gcu') and torch.gcu.is_available():
+            if torch.gcu.is_bf16_supported():
+                bf_16_support = True
+    elif device_name.startswith("musa"):
+        if hasattr(torch, 'musa') and torch.musa.is_available():
+            if torch.musa.is_bf16_supported():
+                bf_16_support = True
+    elif device_name.startswith("npu"):
+        if hasattr(torch, 'npu') and torch.npu.is_available():
+            if torch.npu.is_bf16_supported():
+                bf_16_support = True
 
     if model_name == 'layoutreader':
         # 检测modelscope的缓存目录是否存在

mineru/utils/config_reader.py

@@ -86,7 +86,15 @@ def get_device():
                 if torch_npu.npu.is_available():
                     return "npu"
             except Exception as e:
-                pass
+                try:
+                    if torch.gcu.is_available():
+                        return "gcu"
+                except Exception as e:
+                    try:
+                        if torch.musa.is_available():
+                            return "musa"
+                    except Exception as e:
+                        pass
         return "cpu"
 
 

mineru/utils/model_utils.py

@@ -414,7 +414,7 @@ def get_res_list_from_layout_res(layout_res, iou_threshold=0.7, overlap_threshol
 
 
 def clean_memory(device='cuda'):
-    if device == 'cuda':
+    if str(device).startswith("cuda"):
         if torch.cuda.is_available():
             torch.cuda.empty_cache()
             torch.cuda.ipc_collect()
@@ -423,6 +423,12 @@ def clean_memory(device='cuda'):
             torch_npu.npu.empty_cache()
     elif str(device).startswith("mps"):
         torch.mps.empty_cache()
+    elif str(device).startswith("gcu"):
+        if torch.gcu.is_available():
+            torch.gcu.empty_cache()
+    elif str(device).startswith("musa"):
+        if torch.musa.is_available():
+            torch.musa.empty_cache()
     gc.collect()
 
 
@@ -458,5 +464,11 @@ def get_vram(device) -> int:
     elif str(device).startswith("npu"):
         if torch_npu.npu.is_available():
             total_memory = round(torch_npu.npu.get_device_properties(device).total_memory / (1024 ** 3))  # 转为 GB
+    elif str(device).startswith("gcu"):
+        if torch.gcu.is_available():
+            total_memory = round(torch.gcu.get_device_properties(device).total_memory / (1024 ** 3))  # 转为 GB
+    elif str(device).startswith("musa"):
+        if torch.musa.is_available():
+            total_memory = round(torch.musa.get_device_properties(device).total_memory / (1024 ** 3))  # 转为 GB
 
     return total_memory

mineru/utils/pdf_image_tools.py

@@ -5,7 +5,7 @@ from io import BytesIO
 import numpy as np
 import pypdfium2 as pdfium
 from loguru import logger
-from PIL import Image
+from PIL import Image, ImageOps
 
 from mineru.data.data_reader_writer import FileBasedDataWriter
 from mineru.utils.check_sys_env import is_windows_environment
@@ -41,19 +41,23 @@ def pdf_page_to_image(page: pdfium.PdfPage, dpi=200, image_type=ImageType.PIL) -
     return image_dict
 
 
-def _load_images_from_pdf_worker(pdf_bytes, dpi, start_page_id, end_page_id, image_type):
+def _load_images_from_pdf_worker(
+    pdf_bytes, dpi, start_page_id, end_page_id, image_type
+):
     """用于进程池的包装函数"""
-    return load_images_from_pdf_core(pdf_bytes, dpi, start_page_id, end_page_id, image_type)
+    return load_images_from_pdf_core(
+        pdf_bytes, dpi, start_page_id, end_page_id, image_type
+    )
 
 
 def load_images_from_pdf(
-        pdf_bytes: bytes,
-        dpi=200,
-        start_page_id=0,
-        end_page_id=None,
-        image_type=ImageType.PIL,
-        timeout=None,
-        threads=4,
+    pdf_bytes: bytes,
+    dpi=200,
+    start_page_id=0,
+    end_page_id=None,
+    image_type=ImageType.PIL,
+    timeout=None,
+    threads=4,
 ):
     """带超时控制的 PDF 转图片函数,支持多进程加速
 
@@ -77,7 +81,7 @@ def load_images_from_pdf(
             dpi,
             start_page_id,
             get_end_page_id(end_page_id, len(pdf_doc)),
-            image_type
+            image_type,
         ), pdf_doc
     else:
         if timeout is None:
@@ -116,7 +120,7 @@ def load_images_from_pdf(
                     dpi,
                     range_start,
                     range_end,
-                    image_type
+                    image_type,
                 )
                 futures.append((range_start, future))
 
@@ -163,7 +167,14 @@ def load_images_from_pdf_core(
     return images_list
 
 
-def cut_image(bbox: tuple, page_num: int, page_pil_img, return_path, image_writer: FileBasedDataWriter, scale=2):
+def cut_image(
+    bbox: tuple,
+    page_num: int,
+    page_pil_img,
+    return_path,
+    image_writer: FileBasedDataWriter,
+    scale=2,
+):
     """从第page_num页的page中，根据bbox进行裁剪出一张jpg图片，返回图片路径 save_path：需要同时支持s3和本地,
     图片存放在save_path下，文件名是:
     {page_num}_{bbox[0]}_{bbox[1]}_{bbox[2]}_{bbox[3]}.jpg , bbox内数字取整。"""
@@ -197,7 +208,6 @@ def get_crop_img(bbox: tuple, pil_img, scale=2):
 
 
 def get_crop_np_img(bbox: tuple, input_img, scale=2):
-
     if isinstance(input_img, Image.Image):
         np_img = np.asarray(input_img)
     elif isinstance(input_img, np.ndarray):
@@ -212,17 +222,27 @@ def get_crop_np_img(bbox: tuple, input_img, scale=2):
         int(bbox[3] * scale),
     )
 
-    return np_img[scale_bbox[1]:scale_bbox[3], scale_bbox[0]:scale_bbox[2]]
+    return np_img[scale_bbox[1] : scale_bbox[3], scale_bbox[0] : scale_bbox[2]]
+
 
 def images_bytes_to_pdf_bytes(image_bytes):
     # 内存缓冲区
     pdf_buffer = BytesIO()
 
     # 载入并转换所有图像为 RGB 模式
-    image = Image.open(BytesIO(image_bytes)).convert("RGB")
+    image = Image.open(BytesIO(image_bytes))
+    # 根据 EXIF 信息自动转正（处理手机拍摄的带 Orientation 标记的图片）
+    image = ImageOps.exif_transpose(image) or image
+    # 只在必要时转换
+    if image.mode != "RGB":
+        image = image.convert("RGB")
 
     # 第一张图保存为 PDF，其余追加
-    image.save(pdf_buffer, format="PDF", save_all=True)
+    image.save(
+        pdf_buffer,
+        format="PDF",
+        # save_all=True
+    )
 
     # 获取 PDF bytes 并重置指针（可选）
     pdf_bytes = pdf_buffer.getvalue()

mineru/utils/table_merge.py

@@ -9,7 +9,20 @@ from mineru.utils.char_utils import full_to_half
 from mineru.utils.enum_class import BlockType, SplitFlag
 
 
-CONTINUATION_MARKERS = ["(续)", "(续表)", "(continued)", "(cont.)"]
+CONTINUATION_END_MARKERS = [
+    "(续)",
+    "(续表)",
+    "(续上表)",
+    "(continued)",
+    "(cont.)",
+    "(cont’d)",
+    "(…continued)",
+    "续表",
+]
+
+CONTINUATION_INLINE_MARKERS = [
+    "(continued)",
+]
 
 
 def calculate_table_total_columns(soup):
@@ -57,6 +70,69 @@ def calculate_table_total_columns(soup):
     return max_cols
 
 
+def build_table_occupied_matrix(soup):
+    """构建表格的占用矩阵，返回每行的有效列数
+
+    Args:
+        soup: BeautifulSoup解析的表格
+
+    Returns:
+        dict: {row_idx: effective_columns} 每行的有效列数（考虑rowspan占用）
+    """
+    rows = soup.find_all("tr")
+    if not rows:
+        return {}
+
+    occupied = {}  # {row_idx: {col_idx: True}}
+    row_effective_cols = {}  # {row_idx: effective_columns}
+
+    for row_idx, row in enumerate(rows):
+        col_idx = 0
+        cells = row.find_all(["td", "th"])
+
+        if row_idx not in occupied:
+            occupied[row_idx] = {}
+
+        for cell in cells:
+            # 找到下一个未被占用的列位置
+            while col_idx in occupied[row_idx]:
+                col_idx += 1
+
+            colspan = int(cell.get("colspan", 1))
+            rowspan = int(cell.get("rowspan", 1))
+
+            # 标记被这个单元格占用的所有位置
+            for r in range(row_idx, row_idx + rowspan):
+                if r not in occupied:
+                    occupied[r] = {}
+                for c in range(col_idx, col_idx + colspan):
+                    occupied[r][c] = True
+
+            col_idx += colspan
+
+        # 该行的有效列数为已占用的最大列索引+1
+        if occupied[row_idx]:
+            row_effective_cols[row_idx] = max(occupied[row_idx].keys()) + 1
+        else:
+            row_effective_cols[row_idx] = 0
+
+    return row_effective_cols
+
+
+def calculate_row_effective_columns(soup, row_idx):
+    """计算指定行的有效列数（考虑rowspan占用）
+
+    Args:
+        soup: BeautifulSoup解析的表格
+        row_idx: 行索引
+
+    Returns:
+        int: 该行的有效列数
+    """
+    row_effective_cols = build_table_occupied_matrix(soup)
+    return row_effective_cols.get(row_idx, 0)
+
+
 def calculate_row_columns(row):
     """
     计算表格行的实际列数，考虑colspan属性
@@ -106,6 +182,10 @@ def detect_table_headers(soup1, soup2, max_header_rows=5):
     rows1 = soup1.find_all("tr")
     rows2 = soup2.find_all("tr")
 
+    # 构建两个表格的有效列数矩阵
+    effective_cols1 = build_table_occupied_matrix(soup1)
+    effective_cols2 = build_table_occupied_matrix(soup2)
+
     min_rows = min(len(rows1), len(rows2), max_header_rows)
     header_rows = 0
     headers_match = True
@@ -123,20 +203,24 @@ def detect_table_headers(soup1, soup2, max_header_rows=5):
         if len(cells1) != len(cells2):
             structure_match = False
         else:
-            # 然后检查单元格的属性和内容
-            for cell1, cell2 in zip(cells1, cells2):
-                colspan1 = int(cell1.get("colspan", 1))
-                rowspan1 = int(cell1.get("rowspan", 1))
-                colspan2 = int(cell2.get("colspan", 1))
-                rowspan2 = int(cell2.get("rowspan", 1))
-
-                # 去除所有空白字符（包括空格、换行、制表符等）
-                text1 = ''.join(full_to_half(cell1.get_text()).split())
-                text2 = ''.join(full_to_half(cell2.get_text()).split())
-
-                if colspan1 != colspan2 or rowspan1 != rowspan2 or text1 != text2:
-                    structure_match = False
-                    break
+            # 检查有效列数是否一致（考虑rowspan影响）
+            if effective_cols1.get(i, 0) != effective_cols2.get(i, 0):
+                structure_match = False
+            else:
+                # 然后检查单元格的属性和内容
+                for cell1, cell2 in zip(cells1, cells2):
+                    colspan1 = int(cell1.get("colspan", 1))
+                    rowspan1 = int(cell1.get("rowspan", 1))
+                    colspan2 = int(cell2.get("colspan", 1))
+                    rowspan2 = int(cell2.get("rowspan", 1))
+
+                    # 去除所有空白字符（包括空格、换行、制表符等）
+                    text1 = ''.join(full_to_half(cell1.get_text()).split())
+                    text2 = ''.join(full_to_half(cell2.get_text()).split())
+
+                    if colspan1 != colspan2 or rowspan1 != rowspan2 or text1 != text2:
+                        structure_match = False
+                        break
 
         if structure_match:
             header_rows += 1
@@ -146,7 +230,54 @@ def detect_table_headers(soup1, soup2, max_header_rows=5):
             headers_match = header_rows > 0  # 只有当至少匹配了一行时，才认为表头匹配
             break
 
-    # 如果没有找到匹配的表头行，则返回失败
+    # 如果严格匹配失败，尝试视觉一致性匹配（只比较文本内容）
+    if header_rows == 0:
+        header_rows, headers_match, header_texts = _detect_table_headers_visual(soup1, soup2, rows1, rows2, max_header_rows)
+
+    return header_rows, headers_match, header_texts
+
+
+def _detect_table_headers_visual(soup1, soup2, rows1, rows2, max_header_rows=5):
+    """
+    基于视觉一致性检测表头（只比较文本内容，忽略colspan/rowspan差异）
+
+    Args:
+        soup1: 第一个表格的BeautifulSoup对象
+        soup2: 第二个表格的BeautifulSoup对象
+        rows1: 第一个表格的行列表
+        rows2: 第二个表格的行列表
+        max_header_rows: 最大可能的表头行数
+
+    Returns:
+        tuple: (表头行数, 表头是否一致, 表头文本列表)
+    """
+    # 构建两个表格的有效列数矩阵
+    effective_cols1 = build_table_occupied_matrix(soup1)
+    effective_cols2 = build_table_occupied_matrix(soup2)
+
+    min_rows = min(len(rows1), len(rows2), max_header_rows)
+    header_rows = 0
+    headers_match = True
+    header_texts = []
+
+    for i in range(min_rows):
+        cells1 = rows1[i].find_all(["td", "th"])
+        cells2 = rows2[i].find_all(["td", "th"])
+
+        # 提取每行的文本内容列表（去除空白字符）
+        texts1 = [''.join(full_to_half(cell.get_text()).split()) for cell in cells1]
+        texts2 = [''.join(full_to_half(cell.get_text()).split()) for cell in cells2]
+
+        # 检查视觉一致性：文本内容完全相同，且有效列数一致
+        effective_cols_match = effective_cols1.get(i, 0) == effective_cols2.get(i, 0)
+        if texts1 == texts2 and effective_cols_match:
+            header_rows += 1
+            row_texts = [full_to_half(cell.get_text().strip()) for cell in cells1]
+            header_texts.append(row_texts)
+        else:
+            headers_match = header_rows > 0
+            break
+
     if header_rows == 0:
         headers_match = False
 
@@ -156,20 +287,32 @@ def detect_table_headers(soup1, soup2, max_header_rows=5):
 def can_merge_tables(current_table_block, previous_table_block):
     """判断两个表格是否可以合并"""
     # 检查表格是否有caption和footnote
+    # 计算previous_table_block中的footnote数量
+    footnote_count = sum(1 for block in previous_table_block["blocks"] if block["type"] == BlockType.TABLE_FOOTNOTE)
     # 如果有TABLE_CAPTION类型的块,检查是否至少有一个以"(续)"结尾
     caption_blocks = [block for block in current_table_block["blocks"] if block["type"] == BlockType.TABLE_CAPTION]
     if caption_blocks:
-        # 如果所有caption都不以"(续)"、"(续表)"、"(continued)"或"(cont.)"结尾,则不合并
+        # 检查是否至少有一个caption包含续表标识
+        has_continuation_marker = False
+        for block in caption_blocks:
+            caption_text = full_to_half(merge_para_with_text(block).strip()).lower()
+            if (
+                    any(caption_text.endswith(marker.lower()) for marker in CONTINUATION_END_MARKERS)
+                    or any(marker.lower() in caption_text for marker in CONTINUATION_INLINE_MARKERS)
+            ):
+                has_continuation_marker = True
+                break
 
-        if not any(
-                any(full_to_half(merge_para_with_text(block).strip()).lower().endswith(marker.lower())
-                    for marker in CONTINUATION_MARKERS)
-                for block in caption_blocks
-        ):
+        # 如果所有caption都不包含续表标识，则不允许合并
+        if not has_continuation_marker:
             return False, None, None, None, None
 
-    if any(block["type"] == BlockType.TABLE_FOOTNOTE for block in previous_table_block["blocks"]):
-        return False, None, None, None, None
+        # 如果current_table_block的caption存在续标识,放宽footnote的限制允许previous_table_block有最多一条footnote
+        if footnote_count > 1:
+            return False, None, None, None, None
+    else:
+        if footnote_count > 0:
+            return False, None, None, None, None
 
     # 获取两个表格的HTML内容
     current_html = ""
@@ -219,34 +362,44 @@ def check_rows_match(soup1, soup2):
     if not (rows1 and rows2):
         return False
 
-    # 获取第一个表的最后一行数据行
+    # 获取第一个表的最后一行数据行索引
+    last_row_idx = None
     last_row = None
-    for row in reversed(rows1):
-        if row.find_all(["td", "th"]):
-            last_row = row
+    for idx in range(len(rows1) - 1, -1, -1):
+        if rows1[idx].find_all(["td", "th"]):
+            last_row_idx = idx
+            last_row = rows1[idx]
             break
 
     # 检测表头行数，以便获取第二个表的首个数据行
     header_count, _, _ = detect_table_headers(soup1, soup2)
 
     # 获取第二个表的首个数据行
+    first_data_row_idx = None
     first_data_row = None
     if len(rows2) > header_count:
+        first_data_row_idx = header_count
         first_data_row = rows2[header_count]  # 第一个非表头行
 
     if not (last_row and first_data_row):
         return False
 
-    # 计算实际列数（考虑colspan）和视觉列数
+    # 计算有效列数（考虑rowspan和colspan）
+    last_row_effective_cols = calculate_row_effective_columns(soup1, last_row_idx)
+    first_row_effective_cols = calculate_row_effective_columns(soup2, first_data_row_idx)
+
+    # 计算实际列数（仅考虑colspan）和视觉列数
     last_row_cols = calculate_row_columns(last_row)
     first_row_cols = calculate_row_columns(first_data_row)
     last_row_visual_cols = calculate_visual_columns(last_row)
     first_row_visual_cols = calculate_visual_columns(first_data_row)
 
-    # logger.debug(f"行列数 - 前表最后一行: {last_row_cols}(视觉列数:{last_row_visual_cols}), 当前表首行: {first_row_cols}(视觉列数:{first_row_visual_cols})")
+    # logger.debug(f"行列数 - 前表最后一行: {last_row_cols}(有效列数:{last_row_effective_cols}, 视觉列数:{last_row_visual_cols}), 当前表首行: {first_row_cols}(有效列数:{first_row_effective_cols}, 视觉列数:{first_row_visual_cols})")
 
-    # 同时考虑实际列数匹配和视觉列数匹配
-    return last_row_cols == first_row_cols or last_row_visual_cols == first_row_visual_cols
+    # 同时考虑有效列数匹配、实际列数匹配和视觉列数匹配
+    return (last_row_effective_cols == first_row_effective_cols or
+            last_row_cols == first_row_cols or
+            last_row_visual_cols == first_row_visual_cols)
 
 
 def check_row_columns_match(row1, row2):
@@ -263,12 +416,13 @@ def check_row_columns_match(row1, row2):
     return True
 
 
-def adjust_table_rows_colspan(rows, start_idx, end_idx,
+def adjust_table_rows_colspan(soup, rows, start_idx, end_idx,
                               reference_structure, reference_visual_cols,
                               target_cols, current_cols, reference_row):
     """调整表格行的colspan属性以匹配目标列数
 
     Args:
+        soup: BeautifulSoup解析的表格对象（用于计算有效列数）
         rows: 表格行列表
         start_idx: 起始行索引
         end_idx: 结束行索引（不包含）
@@ -280,14 +434,21 @@ def adjust_table_rows_colspan(rows, start_idx, end_idx,
     """
     reference_row_copy = deepcopy(reference_row)
 
+    # 构建有效列数矩阵
+    effective_cols_matrix = build_table_occupied_matrix(soup)
+
     for i in range(start_idx, end_idx):
         row = rows[i]
         cells = row.find_all(["td", "th"])
         if not cells:
             continue
 
+        # 使用有效列数（考虑rowspan）判断是否需要调整
+        current_row_effective_cols = effective_cols_matrix.get(i, 0)
         current_row_cols = calculate_row_columns(row)
-        if current_row_cols >= target_cols:
+
+        # 如果有效列数或实际列数已经达到目标，则跳过
+        if current_row_effective_cols >= target_cols or current_row_cols >= target_cols:
             continue
 
         # 检查是否与参考行结构匹配
@@ -299,9 +460,12 @@ def adjust_table_rows_colspan(rows, start_idx, end_idx,
                         cell["colspan"] = str(reference_structure[j])
         else:
             # 扩展最后一个单元格以填补列数差异
-            last_cell = cells[-1]
-            current_last_span = int(last_cell.get("colspan", 1))
-            last_cell["colspan"] = str(current_last_span + (target_cols - current_cols))
+            # 使用有效列数来计算差异
+            cols_diff = target_cols - current_row_effective_cols
+            if cols_diff > 0:
+                last_cell = cells[-1]
+                current_last_span = int(last_cell.get("colspan", 1))
+                last_cell["colspan"] = str(current_last_span + cols_diff)
 
 
 def perform_table_merge(soup1, soup2, previous_table_block, wait_merge_table_footnotes):
@@ -332,7 +496,7 @@ def perform_table_merge(soup1, soup2, previous_table_block, wait_merge_table_foo
             reference_visual_cols = calculate_visual_columns(last_row1)
             # 以表1的最后一行为参考，调整表2的行
             adjust_table_rows_colspan(
-                rows2, header_count, len(rows2),
+                soup2, rows2, header_count, len(rows2),
                 reference_structure, reference_visual_cols,
                 table_cols1, table_cols2, first_data_row2
             )
@@ -342,7 +506,7 @@ def perform_table_merge(soup1, soup2, previous_table_block, wait_merge_table_foo
             reference_visual_cols = calculate_visual_columns(first_data_row2)
             # 以表2的第一个数据行为参考，调整表1的行
             adjust_table_rows_colspan(
-                rows1, 0, len(rows1),
+                soup1, rows1, 0, len(rows1),
                 reference_structure, reference_visual_cols,
                 table_cols2, table_cols1, last_row1
             )
@@ -356,6 +520,11 @@ def perform_table_merge(soup1, soup2, previous_table_block, wait_merge_table_foo
                 row.extract()
                 tbody1.append(row)
 
+    # 清空previous_table_block的footnote
+    previous_table_block["blocks"] = [
+        block for block in previous_table_block["blocks"]
+        if block["type"] != BlockType.TABLE_FOOTNOTE
+    ]
     # 添加待合并表格的footnote到前一个表格中
     for table_footnote in wait_merge_table_footnotes:
         temp_table_footnote = table_footnote.copy()
@@ -416,4 +585,4 @@ def merge_table(page_info_list):
         # 删除当前页的table
         for block in current_table_block["blocks"]:
             block['lines'] = []
-            block[SplitFlag.LINES_DELETED] = True
+            block[SplitFlag.LINES_DELETED] = True

mineru/version.py

@@ -1 +1 @@
-__version__ = "2.7.0"
+__version__ = "2.7.2"

{mineru-2.7.0.dist-info → mineru-2.7.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mineru
-Version: 2.7.0
+Version: 2.7.2
 Summary: A practical tool for converting PDF to Markdown
 License: AGPL-3.0
 Project-URL: homepage, https://mineru.net/
@@ -19,7 +19,7 @@ Requires-Dist: boto3>=1.28.43
 Requires-Dist: click>=8.1.7
 Requires-Dist: loguru>=0.7.2
 Requires-Dist: numpy>=1.21.6
-Requires-Dist: pdfminer.six==20250506
+Requires-Dist: pdfminer.six>=20251230
 Requires-Dist: tqdm>=4.67.1
 Requires-Dist: requests
 Requires-Dist: httpx
@@ -60,7 +60,7 @@ Requires-Dist: matplotlib<4,>=3.10; extra == "pipeline"
 Requires-Dist: ultralytics<9,>=8.3.48; extra == "pipeline"
 Requires-Dist: doclayout_yolo==0.0.4; extra == "pipeline"
 Requires-Dist: dill<1,>=0.3.8; extra == "pipeline"
-Requires-Dist: PyYAML<7,>=6.0.2; extra == "pipeline"
+Requires-Dist: PyYAML<7,>=6.0.1; extra == "pipeline"
 Requires-Dist: ftfy<7,>=6.3.1; extra == "pipeline"
 Requires-Dist: shapely<3,>=2.0.7; extra == "pipeline"
 Requires-Dist: pyclipper<2,>=1.3.0; extra == "pipeline"
@@ -81,9 +81,9 @@ Requires-Dist: mineru[vlm]; extra == "core"
 Requires-Dist: mineru[pipeline]; extra == "core"
 Requires-Dist: mineru[api]; extra == "core"
 Requires-Dist: mineru[gradio]; extra == "core"
-Requires-Dist: mineru[mlx]; sys_platform == "darwin" and extra == "core"
 Provides-Extra: all
 Requires-Dist: mineru[core]; extra == "all"
+Requires-Dist: mineru[mlx]; sys_platform == "darwin" and extra == "all"
 Requires-Dist: mineru[vllm]; sys_platform == "linux" and extra == "all"
 Requires-Dist: mineru[lmdeploy]; sys_platform == "windows" and extra == "all"
 Dynamic: license-file
@@ -135,6 +135,22 @@ Dynamic: license-file
 
 # Changelog
 
+- 2026/01/23 2.7.2 Release
+  - Added support for domestic computing platforms Hygon, Enflame, and Moore Threads. Currently, the officially supported domestic computing platforms include:
+    - [Ascend](https://opendatalab.github.io/MinerU/zh/usage/acceleration_cards/Ascend/) 
+    - [T-Head](https://opendatalab.github.io/MinerU/zh/usage/acceleration_cards/THead/) 
+    - [METAX](https://opendatalab.github.io/MinerU/zh/usage/acceleration_cards/METAX/) 
+    - [Hygon](https://opendatalab.github.io/MinerU/zh/usage/acceleration_cards/Hygon/)
+    - [Enflame](https://opendatalab.github.io/MinerU/zh/usage/acceleration_cards/Enflame/)
+    - [MooreThreads](https://opendatalab.github.io/MinerU/zh/usage/acceleration_cards/MooreThreads/)
+  - MinerU continues to ensure compatibility with domestic hardware platforms, supporting mainstream chip architectures. With secure and reliable technology, we empower researchers, government, and enterprises to reach new heights in document digitization!
+  - Cross-page table merging optimization, improving merge success rate and merge quality
+
+- 2026/01/06 2.7.1 Release
+  - fix bug: #4300
+  - Updated pdfminer.six dependency version to resolve [CVE-2025-64512](https://github.com/advisories/GHSA-wf5f-4jwr-ppcp)
+  - Support automatic correction of input image exif orientation to improve OCR recognition accuracy  #4283
+
 - 2025/12/30 2.7.0 Release
   - Simplified installation process. No need to separately install `vlm` acceleration engine dependencies. Using `uv pip install mineru[all]` during installation will install all optional backend dependencies.
   - Added new `hybrid` backend, which combines the advantages of `pipeline` and `vlm` backends. Built on vlm, it integrates some capabilities of pipeline, adding extra extensibility on top of high accuracy:

{mineru-2.7.0.dist-info → mineru-2.7.2.dist-info}/RECORD

@@ -1,5 +1,5 @@
 mineru/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
-mineru/version.py,sha256=EtKWW0Hnl5oWglRNH0HZigvcDT2FEs58ek8buJdwW1E,22
+mineru/version.py,sha256=H1WLrviWKvrPzDle8EWdCYYkzljxs0mtbXigYc-xaKA,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
@@ -17,13 +17,13 @@ mineru/backend/pipeline/pipeline_magic_model.py,sha256=w8jGx8f6yZN0Wf2yPP3L9rYKc
 mineru/backend/pipeline/pipeline_middle_json_mkcontent.py,sha256=NJCLGKE7BqM24bRdpXCfTalyiqozowFZjpdzpIUy5aA,14672
 mineru/backend/vlm/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
 mineru/backend/vlm/model_output_to_middle_json.py,sha256=AqYX44gS9crUO_t7SuUatD71EVjow6pI6yA2Ik3gQ0s,5139
-mineru/backend/vlm/utils.py,sha256=sqcS4WVCcxVL1aElKII1zNYMu2yH5tRpVqRb4lXVm38,3650
-mineru/backend/vlm/vlm_analyze.py,sha256=EdfEmkroA3lafRZLqN4uOaLWx9oxVnUocqFsWZyS82c,11666
+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_middle_json_mkcontent.py,sha256=w-Szbm4HitR7MY4pinSCZZdXtPSqmtlU9cjNh4IOQyg,29499
 mineru/cli/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
 mineru/cli/client.py,sha256=mPNfMEShVG-ithmlJQ5nGRIad2gCZgUjBGHN7zAmLhQ,6978
-mineru/cli/common.py,sha256=srzkJeh0dMlaccDlGWaDBIToqXLPFeyYL4JsARCGHTc,20468
+mineru/cli/common.py,sha256=fMPc235DtnupQkh9uFIMHUpxOSvCp5yc3A56sAabAWY,20475
 mineru/cli/fast_api.py,sha256=TGpZqyUE1kg2eXsP76pr0p1yqNOOU9jyjL5Pc0FJwRc,16637
 mineru/cli/gradio_app.py,sha256=2IIWOm2bEHHq5BZMlfmN3yAJw1Nf8SUALTQ95o-bYy0,21863
 mineru/cli/models_download.py,sha256=LNfoIpUlJM7m7qb2SiCxtjMDw4jILBQtZwNP2JoY81U,4815
@@ -47,13 +47,13 @@ mineru/model/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
 mineru/model/layout/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
 mineru/model/layout/doclayoutyolo.py,sha256=DttINdulzTiYcVDl_70oDtUdfVmGc9qkKWmbPOGAeV0,3867
 mineru/model/mfd/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
-mineru/model/mfd/yolo_v8.py,sha256=t7ptmShHoZCW9WkqLNCL1KRChxpa2E7j5g5fibXlUvY,3681
+mineru/model/mfd/yolo_v8.py,sha256=OI5AxVgt3FvXp4NYk0BDXXvpDlo9YjM6byDyC_TZ8Js,3714
 mineru/model/mfr/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
 mineru/model/mfr/utils.py,sha256=pAi1HnkTuO0R6251Hdl-o50m0wH0Ce89PAf74WCsXPU,11499
 mineru/model/mfr/pp_formulanet_plus_m/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 mineru/model/mfr/pp_formulanet_plus_m/predict_formula.py,sha256=alGX_sPJxZh_7v1sOK3DJ8akfkWO-2c5I_JR7aXMTLU,5588
 mineru/model/mfr/pp_formulanet_plus_m/processors.py,sha256=MSKyanxiDDjgDQHBov-GjKtPnMx9tSmxBC9GIkM3ft8,23832
-mineru/model/mfr/unimernet/Unimernet.py,sha256=1SGLSQ2rc6oslnEwP4Ti7JxaNlyCSGge0js-Tr1VikE,7864
+mineru/model/mfr/unimernet/Unimernet.py,sha256=ZK0M9fPmZziK4D33H3YND7RnHiQkRVCS-lvNfY-N7do,7912
 mineru/model/mfr/unimernet/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 mineru/model/mfr/unimernet/unimernet_hf/__init__.py,sha256=kHcISG8GS4TWJW34SCJCei1jxo6HxvO00aC0dqyNFgI,413
 mineru/model/mfr/unimernet/unimernet_hf/modeling_unimernet.py,sha256=_lN3zDKxeqsW-h9tXx79DYiT5uT4P9ixG49WrSYKFxE,7551
@@ -88,7 +88,7 @@ mineru/model/table/rec/unet_table/main.py,sha256=J13Q7_6stYyedmVedf9CZD7R0tuguGf
 mineru/model/table/rec/unet_table/table_recover.py,sha256=rSyeWyuP10M8dLKA5e0n4P2DXMYbVbmgLxEcdZA8_0E,9059
 mineru/model/table/rec/unet_table/table_structure_unet.py,sha256=hnmYLzZFRlK0Y4gr874G9GaLahcKnNZYNun869FdmH8,8150
 mineru/model/table/rec/unet_table/utils.py,sha256=CYAqJW0wePJk4NAemb8W203N7E32v0ujiWbxanDhd8I,16083
-mineru/model/table/rec/unet_table/utils_table_line_rec.py,sha256=zrCdPwI4M8nu0FEfd7lRJAe0z8kYq3KFbzwElM82USE,11174
+mineru/model/table/rec/unet_table/utils_table_line_rec.py,sha256=6z0jYO6S8wAmfHe5tAyEfzWZIQv8wrn_dRU9GC7oKro,11435
 mineru/model/table/rec/unet_table/utils_table_recover.py,sha256=XksJsY82ZS0kqUnNT-jvaYzxJ3V3svMSzj0puwIau1k,10651
 mineru/model/utils/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
 mineru/model/utils/pytorchocr/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -151,17 +151,17 @@ mineru/model/utils/tools/infer/predict_system.py,sha256=hkegkn6hq2v2zqHVAP615-k-
 mineru/model/utils/tools/infer/pytorchocr_utility.py,sha256=i1PFN-_kefJUUZ4Vk7igs1TU8gfErTDlDXY6-8Uaurw,9323
 mineru/model/vlm/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 mineru/model/vlm/lmdeploy_server.py,sha256=PvxJNcUIKB8VzWMDXeV1t0SHSgz_ULO36ZAzJbppz90,3262
-mineru/model/vlm/vllm_server.py,sha256=w5ddusPbcVaEoWAo_BRjmwv_Ywxrc_bCMRhxihoyykY,2263
+mineru/model/vlm/vllm_server.py,sha256=DtYRAHINYN4qkR2onVMofvANPTkSP6tE4IRY_vZgpiA,3079
 mineru/resources/header.html,sha256=7xrf6bGloR-3ZeTDyA-JvavE_NeRuUDe3p07cEKUXSI,4769
 mineru/resources/fasttext-langdetect/lid.176.ftz,sha256=jzRyz-hzintgmejpmcPL-uDc0VaWqsfXc4qAOdtgPoM,938013
 mineru/utils/__init__.py,sha256=8CRrCQVuExa0BttRFh3Z40lFy2K5jN0sp67KWjOlj5c,50
 mineru/utils/block_pre_proc.py,sha256=uGBmxf2MR9bplTnQI8xHjCI-kj3plRhJr0hcWKidbOQ,9632
-mineru/utils/block_sort.py,sha256=5e1mOLB3W7xu5Y1hmhvGSHPL_aQ41R_4VXcP4vjYAOU,12976
+mineru/utils/block_sort.py,sha256=MmgjZBcmaWssAglzE75VixjtJ_BLNUHO0gvCNQHvlY4,13538
 mineru/utils/boxbase.py,sha256=xnGA1k7hVtTQrreqlJmK-SA3y9edTHgLmGiqGrSXckE,7568
 mineru/utils/char_utils.py,sha256=74T5Ylr5mi1uddAIuJku9Z6sH7vhR7t595_H7qmbu4c,1777
 mineru/utils/check_sys_env.py,sha256=TRjzg4xWyoSGrgv4KaP225A-99xBgLAfZ1cPcGqrBAA,1191
 mineru/utils/cli_parser.py,sha256=4seFAu1kulsYnw6WM2q_cxgEOt2tErZVkI-LNEF_kGw,1445
-mineru/utils/config_reader.py,sha256=IRVWTpBnbnRpck6eXZUKw-fcLt7hon5S4uqWW-RBb1w,4075
+mineru/utils/config_reader.py,sha256=mfulokOzI-33sZy7S-wEVbY3z01PdC7X3761fuhqR3s,4393
 mineru/utils/cut_image.py,sha256=g3m4nfcJNWlxi-P0kpXTtlmspXkMcLCfGwmYuQ-Z2hE,751
 mineru/utils/draw_bbox.py,sha256=FkgppjUzRhN-uxvChdkhHXcDavJEaApMD6qC6qoRwfQ,20292
 mineru/utils/engine_utils.py,sha256=Jmao9-O-sZDzH7vANKEDaY6NJ8tuthKsTr23LFIeBLU,2203
@@ -172,22 +172,22 @@ mineru/utils/hash_utils.py,sha256=UPS_8NRBmVumdyOv16Lmv6Ly2xK8OVDJEe5gG6gKIFk,85
 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/model_utils.py,sha256=6moOQqE5ShHaJKkENXP8BXJA7RCWtOGlYHZ3nidwmZs,18977
+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
 mineru/utils/os_env_config.py,sha256=ZNtkR4KrJW72CeIoTNzGDL6tMKv_hL8nzvWIssGWbqY,842
 mineru/utils/pdf_classify.py,sha256=6DF5pH_9Uq83fsFtp7n4i-OdYQGzoNOV9L0VBUhgBMQ,8078
-mineru/utils/pdf_image_tools.py,sha256=86_xvsGOEde5QGlKz5uJemjoO1upr6n_K7o3lCdyIjQ,7981
+mineru/utils/pdf_image_tools.py,sha256=L2kHKoFaQo4CGjS1d68JACrlBycx6gyCnnFlbBFRKuw,8273
 mineru/utils/pdf_page_id.py,sha256=em966k12CRW4Rj49RGiLB_8ILwkXPBnWRetApax3eTs,400
 mineru/utils/pdf_reader.py,sha256=WeINm5SyWBUXT0wP9lzIbeHs8P6WUIkN6nVL5X4LzG4,3267
 mineru/utils/pdf_text_tool.py,sha256=KEztjfdqsIHHuiTEAMAL7Lr1OS3R7Ur-uTqGiCRjReQ,1364
 mineru/utils/run_async.py,sha256=rPeP4BCZerR8VByRDhiYzfZiahLVqoZEBVAS54dAjNg,1286
 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=zPjDwso28bHEaH9mwlQMtPvwAPcdNfwDK-pinXTf7E8,15129
-mineru-2.7.0.dist-info/licenses/LICENSE.md,sha256=jVa0BUaKrRH4erV2P5AeJ24I2WRv9chIGxditreJ6e0,34524
-mineru-2.7.0.dist-info/METADATA,sha256=NoPOfPd9ZWB153FfrhRLnlqBcNShcZP1WEDiCc8U93k,35263
-mineru-2.7.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-mineru-2.7.0.dist-info/entry_points.txt,sha256=JbtrCPhx1T32s7TONUsteKg-24ZwRT1HSiFtW5jypVw,376
-mineru-2.7.0.dist-info/top_level.txt,sha256=zuGQfZcbsHv4I4oKI9gaKPqEWBFm6xJroKuug2LnKP8,7
-mineru-2.7.0.dist-info/RECORD,,
+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-2.7.0.dist-info → mineru-2.7.2.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (80.9.0)
+Generator: setuptools (80.10.1)
 Root-Is-Purelib: true
 Tag: py3-none-any