doc-page-extractor 0.0.6__tar.gz → 0.0.8__tar.gz

{doc_page_extractor-0.0.6 → doc_page_extractor-0.0.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: doc-page-extractor
-Version: 0.0.6
+Version: 0.0.8
 Summary: doc page extractor can identify text and format in images and return structured data.
 Home-page: https://github.com/Moskize91/doc-page-extractor
 Author: Tao Zeyu
@@ -10,7 +10,6 @@ License-File: LICENSE
 Requires-Dist: opencv-python<5.0,>=4.11.0
 Requires-Dist: pillow<11.0,>=10.3
 Requires-Dist: pyclipper<2.0,>=1.2.0
-Requires-Dist: onnxruntime<2.0,>=1.19.0
 Requires-Dist: numpy<2.0,>=1.24.0
 Requires-Dist: shapely<3.0,>=2.0.0
 Requires-Dist: transformers<5.0,>=4.48.0
@@ -37,10 +36,20 @@ doc page extractor can identify text and format in images and return structured
 pip install doc-page-extractor
 ```
 
+```shell
+pip install onnxruntime==1.21.0
+```
+
 ## Using CUDA
 
 Please refer to the introduction of [PyTorch](https://pytorch.org/get-started/locally/) and select the appropriate command to install according to your operating system.
 
+In addition, replace the command to install `onnxruntime` in the previous article with the following:
+
+```shell
+pip install onnxruntime-gpu==1.21.0
+```
+
 ## Example
 
 ```python
@@ -49,7 +58,7 @@ from doc_page_extractor import DocExtractor
 
 extractor = DocExtractor(
   model_dir_path=model_path, # Folder address where AI model is downloaded and installed
-  device="cpu", # If you want to use CUDA, please change to device="cuda:0".
+  device="cpu", # If you want to use CUDA, please change to device="cuda".
 )
 with Image.open("/path/to/your/image.png") as image:
   result = extractor.extract(

{doc_page_extractor-0.0.6 → doc_page_extractor-0.0.8}/README.md

@@ -12,10 +12,20 @@ doc page extractor can identify text and format in images and return structured
 pip install doc-page-extractor
 ```
 
+```shell
+pip install onnxruntime==1.21.0
+```
+
 ## Using CUDA
 
 Please refer to the introduction of [PyTorch](https://pytorch.org/get-started/locally/) and select the appropriate command to install according to your operating system.
 
+In addition, replace the command to install `onnxruntime` in the previous article with the following:
+
+```shell
+pip install onnxruntime-gpu==1.21.0
+```
+
 ## Example
 
 ```python
@@ -24,7 +34,7 @@ from doc_page_extractor import DocExtractor
 
 extractor = DocExtractor(
   model_dir_path=model_path, # Folder address where AI model is downloaded and installed
-  device="cpu", # If you want to use CUDA, please change to device="cuda:0".
+  device="cpu", # If you want to use CUDA, please change to device="cuda".
 )
 with Image.open("/path/to/your/image.png") as image:
   result = extractor.extract(

{doc_page_extractor-0.0.6 → doc_page_extractor-0.0.8}/doc_page_extractor/clipper.py

@@ -73,6 +73,9 @@ def _size_and_wrapper(rect: Rectangle):
     else:
       widths.append(distance)
 
+  if len(widths) == 0 and len(heights) == 0:
+    return 0.0, 0.0, 0.0, 0.0
+
   width: float = sum(widths) / len(widths)
   height: float = sum(heights) / len(heights)
   max_width: float = width

{doc_page_extractor-0.0.6 → doc_page_extractor-0.0.8}/doc_page_extractor/extractor.py

@@ -1,5 +1,4 @@
 import os
-import torch
 
 from typing import Literal, Iterable
 from pathlib import Path
@@ -34,10 +33,6 @@ class DocExtractor:
     self._yolo: YOLOv10 | None = None
     self._layout: LayoutLMv3ForTokenClassification | None = None
 
-    if self._device.startswith("cuda") and not torch.cuda.is_available():
-      self._device = "cpu"
-      print("Warn: cuda is not available, use cpu instead")
-
   def extract(
       self,
       image: Image,
@@ -83,7 +78,7 @@ class DocExtractor:
       source=source,
       imgsz=1024,
       conf=0.2,
-      device=self._device    # Device to use (e.g., "cuda:0" or "cpu")
+      device=self._device    # Device to use (e.g., "cuda" or "cpu")
     )
     boxes = det_res[0].__dict__["boxes"]
     layouts: list[Layout] = []
@@ -142,6 +137,10 @@ class DocExtractor:
   def _find_matched_layout(self, fragment: OCRFragment, layouts: list[Layout]) -> Layout | None:
     fragment_area = fragment.rect.area
     primary_layouts: list[(Layout, float)] = []
+
+    if fragment_area == 0.0:
+      return None
+
     for layout in layouts:
       area = intersection_area(fragment.rect, layout.rect)
       if area / fragment_area > 0.85:
@@ -174,6 +173,9 @@ class DocExtractor:
       fragment.order = i
 
   def _order_fragments_by_ai(self, width: int, height: int, layouts: list[Layout]):
+    if width == 0 or height == 0:
+      return
+
     layout_model = self._get_layout()
     boxes: list[list[int]] = []
     steps: float = 1000.0 # max value of layoutreader
@@ -273,6 +275,9 @@ class DocExtractor:
     width = right - left
     height = bottom - top
 
+    if width == 0 or height == 0:
+      return
+
     for _left, _top, _right, _bottom in boxes:
       yield (
         (_left - left) / width,

{doc_page_extractor-0.0.6 → doc_page_extractor-0.0.8}/doc_page_extractor/ocr.py

@@ -63,16 +63,21 @@ class OCR:
       text, rank = res
       if is_space_text(text):
         continue
+
+      rect = Rectangle(
+        lt=(box[0][0], box[0][1]),
+        rt=(box[1][0], box[1][1]),
+        rb=(box[2][0], box[2][1]),
+        lb=(box[3][0], box[3][1]),
+      )
+      if rect.area == 0.0:
+        continue
+
       yield OCRFragment(
         order=index,
         text=text,
         rank=rank,
-        rect=Rectangle(
-          lt=(box[0][0], box[0][1]),
-          rt=(box[1][0], box[1][1]),
-          rb=(box[2][0], box[2][1]),
-          lb=(box[3][0], box[3][1]),
-        ),
+        rect=rect,
       )
       index += 1
 

{doc_page_extractor-0.0.6 → doc_page_extractor-0.0.8}/doc_page_extractor/overlap.py

@@ -88,6 +88,9 @@ def regroup_lines(origin_fragments: list[OCRFragment]) -> list[OCRFragment]:
         x2 = max(x2, x)
         y2 = max(y2, y)
 
+    if len(proto_texts_len) == 0:
+      continue
+
     fragments.append(OCRFragment(
       order=min_order,
       text=" ".join(texts),
@@ -115,13 +118,14 @@ def _split_fragments_into_groups(fragments: list[OCRFragment]) -> Generator[list
     if len(group) > 0:
       next_mean_median = (sum_median + median) / (len(group) + 1)
       next_mean_height = (sum_height + height) / (len(group) + 1)
-      deviation_rate = abs(median - next_mean_median) / next_mean_height
 
-      if deviation_rate > max_deviation_rate:
-        yield group
-        group = []
-        sum_height = 0.0
-        sum_median = 0.0
+      if next_mean_height > 0:
+        deviation_rate = abs(median - next_mean_median) / next_mean_height
+        if deviation_rate > max_deviation_rate:
+          yield group
+          group = []
+          sum_height = 0.0
+          sum_median = 0.0
 
     group.append(fragment)
     sum_height += height
@@ -141,7 +145,12 @@ def overlap_rate(polygon1: Polygon, polygon2: Polygon) -> float:
   else:
     overlay_width, overlay_height = _polygon_size(intersection)
     polygon2_width, polygon2_height = _polygon_size(polygon2)
-    return (overlay_width / polygon2_width + overlay_height / polygon2_height) / 2.0
+    if polygon2_width == 0.0 or polygon2_height == 0.0:
+      return 0.0
+    return (
+      overlay_width / polygon2_width +
+      overlay_height / polygon2_height
+    ) / 2.0
 
 def _polygon_size(polygon: Polygon) -> tuple[float, float]:
   x1: float = float("inf")

{doc_page_extractor-0.0.6 → doc_page_extractor-0.0.8}/doc_page_extractor/rotation.py

@@ -129,7 +129,9 @@ def _find_median(rotations: list[float]):
     mid2 = rotations[n // 2]
     return (mid1 + mid2) / 2
 
-def _find_mean(rotations: list[float]):
+def _find_mean(rotations: list[float]) -> float:
+  if len(rotations) == 0:
+    return 0.0
   return sum(rotations) / len(rotations)
 
 # rotation is in [0, pi)

{doc_page_extractor-0.0.6 → doc_page_extractor-0.0.8}/doc_page_extractor.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: doc-page-extractor
-Version: 0.0.6
+Version: 0.0.8
 Summary: doc page extractor can identify text and format in images and return structured data.
 Home-page: https://github.com/Moskize91/doc-page-extractor
 Author: Tao Zeyu
@@ -10,7 +10,6 @@ License-File: LICENSE
 Requires-Dist: opencv-python<5.0,>=4.11.0
 Requires-Dist: pillow<11.0,>=10.3
 Requires-Dist: pyclipper<2.0,>=1.2.0
-Requires-Dist: onnxruntime<2.0,>=1.19.0
 Requires-Dist: numpy<2.0,>=1.24.0
 Requires-Dist: shapely<3.0,>=2.0.0
 Requires-Dist: transformers<5.0,>=4.48.0
@@ -37,10 +36,20 @@ doc page extractor can identify text and format in images and return structured
 pip install doc-page-extractor
 ```
 
+```shell
+pip install onnxruntime==1.21.0
+```
+
 ## Using CUDA
 
 Please refer to the introduction of [PyTorch](https://pytorch.org/get-started/locally/) and select the appropriate command to install according to your operating system.
 
+In addition, replace the command to install `onnxruntime` in the previous article with the following:
+
+```shell
+pip install onnxruntime-gpu==1.21.0
+```
+
 ## Example
 
 ```python
@@ -49,7 +58,7 @@ from doc_page_extractor import DocExtractor
 
 extractor = DocExtractor(
   model_dir_path=model_path, # Folder address where AI model is downloaded and installed
-  device="cpu", # If you want to use CUDA, please change to device="cuda:0".
+  device="cpu", # If you want to use CUDA, please change to device="cuda".
 )
 with Image.open("/path/to/your/image.png") as image:
   result = extractor.extract(

{doc_page_extractor-0.0.6 → doc_page_extractor-0.0.8}/doc_page_extractor.egg-info/requires.txt

@@ -1,7 +1,6 @@
 opencv-python<5.0,>=4.11.0
 pillow<11.0,>=10.3
 pyclipper<2.0,>=1.2.0
-onnxruntime<2.0,>=1.19.0
 numpy<2.0,>=1.24.0
 shapely<3.0,>=2.0.0
 transformers<5.0,>=4.48.0

{doc_page_extractor-0.0.6 → doc_page_extractor-0.0.8}/setup.py

@@ -2,7 +2,7 @@ from setuptools import setup, find_packages
 
 setup(
   name="doc-page-extractor",
-  version="0.0.6",
+  version="0.0.8",
   author="Tao Zeyu",
   author_email="i@taozeyu.com",
   url="https://github.com/Moskize91/doc-page-extractor",
@@ -14,7 +14,6 @@ setup(
     "opencv-python>=4.11.0,<5.0",
     "pillow>=10.3,<11.0",
     "pyclipper>=1.2.0,<2.0",
-    "onnxruntime>=1.19.0,<2.0",
     "numpy>=1.24.0,<2.0",
     "shapely>=2.0.0,<3.0",
     "transformers>=4.48.0,<5.0",