pyconverters-openai_vision 0.5.52__tar.gz → 0.5.54__tar.gz

{pyconverters_openai_vision-0.5.52 → pyconverters_openai_vision-0.5.54}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pyconverters-openai_vision
-Version: 0.5.52
+Version: 0.5.54
 Summary: OpenAIVision converter
 Home-page: https://kairntech.com/
 Author: Olivier Terrier

{pyconverters_openai_vision-0.5.52 → pyconverters_openai_vision-0.5.54}/pyconverters_openai_vision/__init__.py

@@ -1,2 +1,2 @@
 """OpenAIVision converter"""
-__version__ = "0.5.52"
+__version__ = "0.5.54"

{pyconverters_openai_vision-0.5.52 → pyconverters_openai_vision-0.5.54}/pyconverters_openai_vision/openai_vision.py

@@ -4,14 +4,14 @@ import re
 from enum import Enum
 from logging import Logger
 from re import Pattern
-from typing import List, cast, Type, Dict, Any, Optional
+from typing import List, cast, Type, Dict, Any, Optional, Callable
 
 import filetype as filetype
 from log_with_context import add_logging_context
 from pydantic import Field, BaseModel
 from pymultirole_plugins.v1.converter import ConverterParameters, ConverterBase
 from pymultirole_plugins.v1.processor import ProcessorParameters, ProcessorBase
-from pymultirole_plugins.v1.schema import Document, AltText
+from pymultirole_plugins.v1.schema import Document, AltText, Sentence
 from starlette.datastructures import UploadFile
 
 from .openai_utils import create_openai_model_enum, openai_chat_completion, gpt_filter, \
@@ -357,6 +357,73 @@ class OpenAIVisionProcessorBaseParameters(ProcessorParameters):
     )
 
 
+def regex_sub_preserve_spans(
+    text: str,
+    regex: str,
+    repl: Callable[[re.Match], str],
+    spans: List[Sentence],
+    flags=0,
+):
+    new_text_parts = []
+    char_map = {}  # old_char_offset -> new_char_offset
+
+    last_pos = 0
+    new_pos = 0
+
+    for match in re.finditer(regex, text, flags):
+        start, end = match.start(), match.end()
+        replacement = repl(match)
+
+        # Copier le texte inchangé
+        unchanged = text[last_pos:start]
+        new_text_parts.append(unchanged)
+
+        for i in range(last_pos, start):
+            char_map[i] = new_pos
+            new_pos += 1
+
+        # Insérer le remplacement
+        new_text_parts.append(replacement)
+
+        for i in range(start, end):
+            char_map[i] = new_pos
+
+        new_pos += len(replacement)
+        last_pos = end
+
+    # Reste du texte
+    tail = text[last_pos:]
+    new_text_parts.append(tail)
+
+    for i in range(last_pos, len(text)):
+        char_map[i] = new_pos
+        new_pos += 1
+
+    new_text = "".join(new_text_parts)
+
+    # Créer le nouveau Doc
+    # Recréer les spans
+    new_spans = None
+    if spans is not None:
+        new_spans = []
+        for span in spans:
+            if span.start not in char_map or span.end - 1 not in char_map:
+                continue
+
+            new_start = char_map[span.start]
+            new_end = char_map[span.end - 1] + 1
+
+            new_span = Sentence(
+                start=new_start,
+                end=new_end,
+                metadata=span.metadata)
+
+            if new_span is not None:
+                new_spans.append(new_span)
+
+    return new_text, new_spans
+
+
 class OpenAIVisionProcessorBase(ProcessorBase):
     __doc__ = """Generate text using [OpenAI Text Completion](https://platform.openai.com/docs/guides/completion) API
     You input some text as a prompt, and the model will generate a text completion that attempts to match whatever context or pattern you gave it."""
@@ -449,20 +516,27 @@ class OpenAIVisionProcessorBase(ProcessorBase):
 
                             if params.replace_refs_altTexts_by_descriptions:
                                 text = document.text
-                                link_regex = r"!\[([^]]+)\]\(([^]]+)\)"
+                                link_regex = r"!\[([^]]+)\]\(([^)]+)\)"
 
                                 def convert_links(matchobj):
                                     m = matchobj.group(0)
                                     m_id = matchobj.group(1)
                                     if m_id in alts:
                                         # markdown blockquote
-                                        m_desc = "\n".join(["> " + li for li in alts[m_id].splitlines()])
+                                        # m_desc = "\n".join(["> " + li for li in alts[m_id].splitlines()])
+                                        m_desc = alts[m_id]
                                         return f"{m}\n{m_desc}\n"
                                     return m
 
-                                ptext = re.sub(link_regex, convert_links, text, 0,
-                                               re.MULTILINE)
-                                document.text = ptext
+                                new_text, new_sentences = regex_sub_preserve_spans(
+                                    text,
+                                    link_regex,
+                                    convert_links,
+                                    document.sentences,
+                                    flags=re.MULTILINE
+                                )
+                                document.text = new_text
+                                document.sentences = new_sentences
                                 for altText in altTexts:
                                     if altText.name not in alts:
                                         document.altTexts.append(altText)

{pyconverters_openai_vision-0.5.52 → pyconverters_openai_vision-0.5.54}/setup.py

@@ -48,7 +48,7 @@ entry_points = \
                           'pyconverters_openai_vision.openai_vision:OpenAIVisionProcessor']}
 
 setup(name='pyconverters-openai_vision',
-      version='0.5.52',
+      version='0.5.54',
       description='OpenAIVision converter',
       author='Olivier Terrier',
       author_email='olivier.terrier@kairntech.com',

{pyconverters_openai_vision-0.5.52 → pyconverters_openai_vision-0.5.54}/tests/data/ENG product fact files_general offer_2025_30pages_alts.json

@@ -132,83 +132,83 @@
   "altTexts": [
     {
       "name": "img-0.jpeg",
-      "text": "The image features a rhythmic gymnast in mid-performance, showcasing a dynamic pose. The gymnast is wearing a sparkling, intricately designed costume that is predominantly pink and adorned with sequins and embellishments. The outfit has a fitted bodice and a flared skirt, enhancing the elegance of the movement. \n\nInstead of a head, there is a large, shiny pink ball positioned where the head would typically be, creating a surreal and whimsical effect. The gymnast's arms are gracefully extended, and one leg is lifted, emphasizing the fluidity and artistry of the routine. The background is a soft, neutral color, which helps to highlight the vibrant costume and the unique composition of the image."
+      "text": "The image depicts a rhythmic gymnast in mid-performance, showcasing a dynamic pose. The gymnast is wearing a sparkling, intricately designed leotard that features a blend of pink and nude colors, adorned with shimmering sequins and embellishments. The outfit highlights the athlete's form and movement. Instead of a head, there is a large, shiny pink ball positioned where the head would typically be, creating a surreal and whimsical effect. The background is a soft, neutral color, emphasizing the gymnast's graceful posture and the vibrant details of the costume. The overall composition captures the elegance and artistry of rhythmic gymnastics."
     },
     {
       "name": "img-1.jpeg",
-      "text": "The image features a textual layout with a bold, large font displaying the year \"2025\" at the top. Below it, in a slightly smaller font, the words \"GENERAL S\" are presented, likely indicating the beginning of a title or section. At the bottom, the phrase \"PRODUCT FACT FIL\" appears, suggesting that this is part of a document or informational sheet related to product details. The background is divided into two colors: a dark section behind the text and a light blue area below."
+      "text": "The image features a textual layout with a black background at the top displaying the year \"2025\" in large, bold, white font. Below that, in a slightly smaller font, the words \"GENERAL S\" are presented, also in white. The bottom section of the image contains the phrase \"PRODUCT FACT FIL\" in a clear, bold font, likely indicating that it is part of a product information document. The overall design is clean and modern, with a light blue section visible at the bottom."
     },
     {
       "name": "img-2.jpeg",
-      "text": "The image presents a pie chart illustrating the distribution of stories published per day in various languages, totaling 292 stories (excluding updates). \n\n- The largest segment, colored blue, represents Arabic, accounting for 35% of the total.\n- The second largest segment, in red, indicates English stories, making up 20%.\n- The orange segment represents Spanish, contributing 14%.\n- The purple segment shows French stories at 16%.\n- The light blue segment indicates German, which comprises 10%.\n- The smallest segment, colored yellow, represents Portuguese, at 5%.\n\nThe chart visually emphasizes the varying proportions of stories across these languages."
+      "text": "The image presents a pie chart illustrating the distribution of stories published per day in various languages, totaling 292 stories (excluding updates). \n\n- The largest segment, colored blue, represents Arabic, accounting for 35% of the total.\n- The second largest segment, in red, indicates English stories, making up 20%.\n- The orange segment represents Spanish, contributing 14%.\n- The purple segment shows French stories at 16%.\n- The light blue segment indicates German, comprising 10%.\n- Finally, the smallest segment, colored yellow, represents Portuguese, which accounts for 5%.\n\nThe chart visually emphasizes the varying proportions of stories across these languages."
     },
     {
       "name": "img-3.jpeg",
-      "text": "The image is a pie chart displaying the distribution of various sports. The largest segment, colored blue, represents \"Football,\" accounting for 57% of the total. Other segments include:\n\n- \"Athletics\" (light blue) at 6%\n- \"Basketball\" (black) at 7%\n- \"Cycling\" (gray) at 3%\n- \"Handball\" (yellow) at 6%\n- \"Motor sports\" (purple) at 6%\n- \"Ski\" (light purple) at 6%\n- \"Tennis\" (green) at 5%\n- \"Olympics\" (orange) at 2%\n\nEach segment is labeled with its corresponding percentage, providing a clear visual representation of the popularity or participation rates of these sports."
+      "text": "The image is a pie chart displaying the distribution of various sports. The largest segment, colored blue, represents \"Football,\" accounting for 57% of the total. Other segments include:\n\n- \"Athletics\" (light blue) at 2%\n- \"Basketball\" (black) at 7%\n- \"Cycling\" (gray) at 3%\n- \"Handball\" (yellow) at 6%\n- \"Motor sports\" (purple) at 6%\n- \"Ski\" (light purple) at 6%\n- \"Tennis\" (green) at 6%\n- \"Olympics\" (orange) at 8%\n\nEach segment is labeled with its corresponding percentage, and the chart uses a variety of colors to differentiate between the sports."
     },
     {
       "name": "img-4.jpeg",
-      "text": "The image presents a pie chart alongside a textual description. The chart illustrates the distribution of daily reports (30 reports per day, excluding updates) across various sports categories. \n\n- The largest segment, colored blue, represents Football, accounting for 67% of the reports.\n- The next largest segment, in purple, indicates Athletics at 9%.\n- Following that, there are smaller segments for Basketball (8%, light blue), Motor Sports (7%, orange), and Tennis (7%, red).\n- The smallest segment, colored yellow, represents the Olympics at 2%.\n\nEach category is clearly labeled with its corresponding color for easy identification."
+      "text": "The image presents a pie chart alongside a textual description. The chart illustrates the distribution of daily reports (excluding updates) across various sports, totaling over 30 reports per day. \n\n- The largest segment, colored blue, represents Football, accounting for 67% of the reports.\n- The next largest segment, in orange, represents Motor Sports at 9%.\n- Following that, the purple segment for Athletics makes up 8%.\n- The light blue segment for Basketball comprises 7%.\n- The red segment for Tennis also accounts for 7%.\n- Finally, the yellow segment for the Olympics represents 2%.\n\nEach segment is labeled with its respective percentage, and a legend identifies the colors corresponding to each sport."
     },
     {
       "name": "img-5.jpeg",
-      "text": "The image presents a visual representation of the distribution of photos taken per day across various sports. The central pie chart is divided into segments, each representing a different sport, with percentages indicating the proportion of total photos attributed to each category. \n\n- The largest segment, colored in blue, represents Football, accounting for 39% of the total.\n- Next is Athletics, shown in orange, making up 22%.\n- The grey segment represents Basketball at 7%.\n- Other sports include Motor Sports (11%, purple), Cricket (4%, red), Rugby (3%, green), and Tennis (3%, light blue).\n- The smallest segments are for Cycling (2%, pink) and the Olympics (9%, light purple).\n\nTo the left of the pie chart, the text states \"1,490* photos per day,\" emphasizing the total number of photos captured daily across these sports."
+      "text": "The image presents a visual representation of the distribution of photos taken per day across various sports. The central pie chart is divided into segments, each representing a different sport, with percentages indicating the proportion of total photos attributed to each category. \n\n- The largest segment, colored blue, represents Football, accounting for 39% of the total.\n- The second largest, in gray, is Basketball, making up 22%.\n- Other segments include Athletics (orange, 7%), Cricket (red, 4%), Cycling (pink, 2%), Motor Sports (purple, 11%), Rugby (green, 3%), Tennis (light blue, 9%), and the Olympics (light purple, 3%).\n\nTo the left of the pie chart, the text states \"1,490* photos per day,\" emphasizing the total number of photos captured daily across these sports."
     },
     {
       "name": "img-6.jpeg",
-      "text": "The image features a simple design with a bright yellow background. In the foreground, the text \"AFP\" is displayed in bold blue letters, followed by a blue circle, and then the word \"Sports\" in a bold black font. The overall layout is clean and straightforward, emphasizing the brand name and its focus on sports."
+      "text": "The image features a simple design with a bright yellow background. In the center, the text \"AFP\" is displayed in bold blue letters, followed by a blue circle. Next to it, the word \"Sports\" is written in a bold black font. The overall layout is clean and straightforward, emphasizing the \"AFP\" branding alongside the sports theme."
     },
     {
       "name": "img-7.jpeg",
-      "text": "The image features a pie chart illustrating the distribution of infographics created per day in different languages. The chart is divided into five segments, each representing a language with a corresponding percentage:\n\n- **German**: 24% (highlighted in yellow)\n- **English**: 24% (highlighted in purple)\n- **Spanish**: 21% (highlighted in light blue)\n- **French**: 16% (highlighted in orange)\n- **Portuguese**: 15% (highlighted in blue)\n\nTo the left of the pie chart, there is a text stating \"6* Infographics per day,\" indicating the total number of infographics produced daily. The overall design is clean and colorful, making it easy to compare the proportions of each language visually."
+      "text": "The image presents a pie chart illustrating the distribution of infographics created per day in different languages. The chart is divided into five segments, each representing a language:\n\n- **German**: 24% (yellow segment)\n- **English**: 24% (purple segment)\n- **Spanish**: 21% (light blue segment)\n- **French**: 16% (orange segment)\n- **Portuguese**: 15% (blue segment)\n\nTo the left of the pie chart, there is a bold text stating \"6* Infographics per day,\" indicating the total number of infographics produced daily. The colors of the segments correspond to the languages listed in the legend next to the chart."
     },
     {
       "name": "img-8.jpeg",
-      "text": "The image features a pie chart illustrating the distribution of videographics produced per year, totaling 75. The chart is divided into five segments, each representing a different language:\n\n- **German**: A small segment colored light blue, accounting for 4% of the total.\n- **English**: A segment colored orange, representing 25%.\n- **Spanish**: The largest segment, colored blue, making up 28%.\n- **French**: A yellow segment, contributing 16%.\n- **Portuguese**: A red segment, which comprises 27%.\n\nAbove the pie chart, the text highlights the total number of videographics produced annually, emphasizing the figure \"75\" in a bold yellow font."
+      "text": "The image features a pie chart illustrating the distribution of videographics produced per year, totaling 75. Each segment of the pie chart is color-coded to represent different languages:\n\n- **German**: Light blue segment, accounting for 4% of the total.\n- **English**: Orange segment, representing 27%.\n- **Spanish**: Dark blue segment, making up 28%.\n- **French**: Yellow segment, which comprises 16%.\n- **Portuguese**: Red segment, contributing 25%.\n\nThe chart visually conveys the proportions of videographics created in each language, highlighting Spanish and English as the most significant contributors. The title above the chart states \"75* videographics per year\" in bold yellow text."
     },
     {
       "name": "img-9.jpeg",
-      "text": "The image features a series of text boxes on a black background, each highlighting key features of a service. \n\n1. The first box states \"Coverage\" followed by \"24 hours a day\" and \"7 days a week,\" indicating continuous availability.\n2. The second box mentions \"+ 300* editable videos per month,\" suggesting a high volume of video content that can be customized.\n3. The third box lists \"130 video production points worldwide,\" emphasizing a global reach for video production services.\n\nThe text is presented in a clean, modern font, enhancing readability against the dark background."
+      "text": "The image features a series of text boxes on a black background, each highlighting different service offerings. \n\n1. The first box states \"Coverage\" followed by \"24 hours a day\" and \"7 days a week,\" indicating continuous availability.\n2. The second box mentions \"+ 300* editable videos per month,\" suggesting a high volume of video content that can be customized.\n3. The third box lists \"130 video production points worldwide,\" emphasizing a global reach for video production services.\n\nThe text is presented in a clean, modern font, enhancing readability against the dark background."
     },
     {
       "name": "img-10.jpeg",
-      "text": "The image features a sleek, modern design with a black background. It presents three key pieces of information in bold, white text, each contained within separate rectangular sections. \n\n1. The first section highlights \"Coverage\" with the details \"24 hours a day\" and \"7 days a week,\" emphasizing continuous availability.\n2. The second section states \"3 Thematic sections,\" indicating a structured approach to content organization.\n3. The third section notes \"+ 15* articles illustrated per day,\" suggesting a high volume of visually supported content produced daily.\n\nThe overall layout is clean and visually appealing, with a focus on clarity and impact."
+      "text": "The image features a sleek, modern design with a black background. It presents three key pieces of information in bold, white text, each contained within separate sections. \n\n1. The first section highlights \"Coverage\" with the details \"24 hours a day\" and \"7 days a week,\" emphasizing continuous availability.\n2. The second section states \"3 Thematic sections,\" indicating a structured approach to content organization.\n3. The final section notes \"+ 15* articles illustrated per day,\" suggesting a high volume of visually supported content produced daily.\n\nThe overall layout is clean and visually appealing, with a focus on clarity and impact."
     },
     {
       "name": "img-11.jpeg",
-      "text": "The image is a pie chart divided into eight segments, each representing a different percentage of a whole. The segments are colored in various shades, including blue, orange, green, purple, yellow, red, and light blue. The percentages displayed on each segment are as follows:\n\n- Light blue: 32%\n- Yellow: 17%\n- Red: 14%\n- Dark blue: 12%\n- Purple: 11%\n- Green: 7%\n- Orange: 7%\n\nThe segments vary in size according to their respective percentages, with the light blue segment being the largest and the orange and green segments being the smallest."
+      "text": "The image depicts a pie chart divided into eight segments, each representing a different percentage of a whole. The segments are colored in various shades: \n\n- The largest segment, colored light blue, occupies 32% of the chart.\n- Next is a yellow segment at 17%.\n- A red segment follows, representing 14%.\n- The blue segment accounts for 12%.\n- There are two segments, both orange and green, each representing 7%.\n- Lastly, a purple segment makes up 11%.\n\nEach segment is labeled with its corresponding percentage, providing a clear visual representation of the data distribution."
     },
     {
       "name": "img-12.jpeg",
-      "text": "The image features a sleek, modern design with a black background. It contains four distinct sections, each highlighted with bold white text. \n\n1. The first section states \"Coverage\" followed by \"24 hours a day\" and \"7 days a week,\" emphasizing continuous availability.\n2. The second section simply displays the number \"5\" alongside the phrase \"Thematic sections,\" indicating a structured approach to content organization.\n3. The third section notes \"+ 60*\" with the phrase \"articles illustrated per day,\" suggesting a high volume of daily content production.\n\nThe overall layout is clean and visually appealing, with a focus on clarity and impact."
+      "text": "The image features a sleek, modern design with a black background. It contains four distinct sections, each highlighted with white text. \n\n1. The first section states \"Coverage\" followed by \"24 hours a day\" and \"7 days a week,\" emphasizing continuous availability.\n2. The second section simply displays the number \"5\" alongside the phrase \"Thematic sections,\" indicating a structured approach to content organization.\n3. The third section notes \"+ 60*\" with the phrase \"articles illustrated per day,\" suggesting a high volume of daily content production.\n\nThe overall layout is clean and visually appealing, with a focus on clarity and impact."
     },
     {
       "name": "img-13.jpeg",
-      "text": "The image is a pie chart divided into several segments, each representing a different percentage of a whole. The segments are colored differently and labeled with their respective percentages. \n\n- The largest segment, colored blue, occupies 35% of the chart.\n- Next, there are segments in purple (15%), black (9%), and two segments in light blue and gray, each at 10%.\n- Other smaller segments include orange (8%), yellow (5%), and two segments in green and another shade, each at 3%.\n\nThe overall design is circular, with clear divisions between the segments, making it easy to visualize the distribution of values represented."
+      "text": "The image is a pie chart divided into several segments, each representing a different percentage of a whole. The largest segment, colored blue, occupies 35% of the chart. Other segments include:\n\n- A purple segment at 15%\n- Two segments at 10% each, one in dark gray and the other in light blue\n- A segment in orange at 8%\n- A yellow segment at 5%\n- Two smaller segments, each at 3%, one in green and the other in black\n- A final segment at 9%, colored gray\n\nThe percentages are displayed prominently within each segment, providing a clear visual representation of the data distribution."
     },
     {
       "name": "img-14.jpeg",
-      "text": "The image features a clean and modern design with three distinct sections, each presented on a black background. \n\n1. The first section states \"Coverage\" followed by \"6am to midnight,\" indicating the hours of operation or coverage.\n2. The second section highlights \"6 Thematic sections,\" suggesting that the content is organized into six different categories or themes.\n3. The third section notes \"+ 45* articles illustrated per day,\" implying a high volume of illustrated articles produced daily.\n\nThe text is presented in a bold, white font, creating a strong contrast against the black background, making it easily readable."
+      "text": "The image features a clean and modern design with a black background. It presents three key pieces of information in bold, white text, each enclosed in a rectangular box. \n\n1. The first box states \"Coverage\" followed by \"6am to midnight,\" indicating the hours of operation or service.\n2. The second box highlights \"6 Thematic sections,\" suggesting a structured approach to content organization.\n3. The third box notes \"+ 45* articles illustrated per day,\" implying a high volume of daily content production, with an asterisk possibly indicating additional context or a footnote.\n\nThe overall layout is simple and visually appealing, emphasizing clarity and accessibility of information."
     },
     {
       "name": "img-15.jpeg",
-      "text": "The image is a pie chart displaying the distribution of various sports. Each segment of the chart is color-coded and labeled with a percentage representing its share of the total.\n\n- **Football** occupies the largest portion at **31%**, represented in blue.\n- **Athletics** follows with **21%**, shown in light blue.\n- **Motor sports** accounts for **17%**, depicted in yellow.\n- **Olympics** makes up **14%**, illustrated in purple.\n- **Basketball** and **Cycling** each represent **6%**, shown in pink and gray, respectively.\n- **Tennis** has the smallest share at **5%**, represented in orange.\n\nThe chart effectively visualizes the popularity or participation rates of these sports."
+      "text": "The image is a pie chart displaying the distribution of various sports. Each segment is color-coded and labeled with a percentage representing its share. \n\n- The largest segment, colored blue, represents \"Football\" at 31%.\n- The second largest, in purple, is \"Athletics\" at 21%.\n- Next, \"Motor sports\" is shown in yellow at 14%.\n- \"Basketball\" is represented in pink at 6%.\n- \"Cycling\" and \"Tennis\" each occupy 6% and are shown in gray and light blue, respectively.\n- Finally, the \"Olympics\" segment is colored orange and accounts for 5%.\n\nThe chart effectively visualizes the popularity or participation rates of these sports."
     },
     {
       "name": "img-16.jpeg",
-      "text": "The image features a clean and modern design with a black background. It contains three distinct sections, each highlighted with white text. \n\n1. The first section states \"Coverage 6am to midnight,\" indicating the hours of operation or availability.\n2. The second section simply states \"2 Thematic sections,\" suggesting that the content is organized into two main categories or themes.\n3. The third section notes \"+ 15* articles illustrated per day,\" implying a high volume of visual content produced daily.\n\nOverall, the layout is straightforward, emphasizing clarity and easy readability."
+      "text": "The image features a minimalist design with a black background and white text. It presents three key points, each enclosed in a rectangular box. \n\n1. The first box states \"Coverage 6am to midnight,\" indicating the operational hours.\n2. The second box highlights \"2 Thematic sections,\" suggesting a structured approach to content organization.\n3. The third box notes \"+ 15* articles illustrated per day,\" emphasizing the volume of illustrated articles produced daily.\n\nThe overall layout is clean and straightforward, focusing on clarity and easy readability."
     },
     {
       "name": "img-17.jpeg",
-      "text": "The image is a pie chart divided into several segments, each representing a percentage of a whole. The largest segment, colored blue, occupies 62% of the chart. The next largest segment is red, representing 15%. Following that, there is a pink segment at 12%. Smaller segments include an orange section at 9%, and three tiny segments colored yellow, light blue, and purple, each representing 1%. The percentages are labeled within or near their respective segments, providing a clear visual representation of the data distribution."
+      "text": "The image is a pie chart divided into several segments, each representing a percentage of a whole. The largest segment, colored blue, occupies 62% of the chart. The next largest segment is red, representing 15%. Following that, there is a pink segment at 12%. Smaller segments include an orange section at 9%, and three tiny segments colored yellow, light blue, and purple, each representing 1%. The percentages are clearly labeled within each segment, providing a visual representation of the data distribution."
     },
     {
       "name": "img-18.jpeg",
-      "text": "The image features a minimalist design with a dark background and white text. It presents three key points related to coverage and content:\n\n1. **Coverage**: The text states \"6am to midnight,\" indicating the hours of operation or availability.\n2. **Thematic Sections**: The number \"1\" is highlighted, suggesting there is a single thematic section.\n3. **Articles**: The phrase \"+10* articles illustrated per day\" implies that more than ten articles are illustrated daily, with an asterisk possibly indicating additional information or conditions related to this point.\n\nOverall, the layout is clean and straightforward, focusing on conveying essential information efficiently."
+      "text": "The image features a minimalist design with a black background and white text. It presents three key points related to coverage and content:\n\n1. **Coverage**: The text states \"6am to midnight,\" indicating the hours of operation or availability.\n2. **Thematic Sections**: The number \"1\" is prominently displayed, suggesting there is one main thematic section.\n3. **Articles**: The phrase \"+ 10* articles illustrated per day\" implies that more than ten articles are illustrated daily, emphasizing the volume of content produced.\n\nOverall, the layout is clean and straightforward, focusing on conveying essential information efficiently."
     },
     {
       "name": "img-19.jpeg",
-      "text": "The image is a pie chart representing the distribution of various sports. The largest segment, colored blue, accounts for 56% and is labeled as \"Football.\" \n\nOther segments include:\n\n- Orange (12%) for \"Basketball\"\n- Red (11%) for \"Tennis\"\n- Yellow (8%) for \"Motor sports\"\n- Light green (10%) for \"Olympics\"\n- Purple (3%) for \"Athletics\"\n\nEach segment is clearly labeled with its corresponding percentage, providing a visual comparison of the popularity or representation of these sports."
+      "text": "The image is a pie chart displaying the distribution of various sports. The largest segment, colored blue, represents \"Football,\" accounting for 56% of the total. Other segments include:\n\n- \"Athletics\" (purple) at 3%\n- \"Basketball\" (orange) at 12%\n- \"Motor sports\" (yellow) at 10%\n- \"Tennis\" (red) at 11%\n- \"Olympics\" (green) at 8%\n\nEach segment is labeled with its corresponding percentage, providing a clear visual representation of the proportions of interest in these sports."
     }
   ]
 }

{pyconverters_openai_vision-0.5.52 → pyconverters_openai_vision-0.5.54}/tests/test_openai_vision.py

@@ -21,6 +21,32 @@ def test_openai_vision_basic():
     assert model_class == OpenAIVisionParameters
 
 
+@pytest.mark.skip(reason="Not a test")
+def test_openai_horrible():
+    start_time = time.time()
+    testdir = Path(__file__).parent
+    parameters = OpenAIVisionProcessorParameters(model_str="gpt-4o", replace_refs_altTexts_by_descriptions=True, prompt="""If the image is a graph (line graph, pie chart, bar chart, etc.), provide a detailed description, possibly a markdown table with figures and headers. Otherwise, just provide a brief one-line description.""")
+    processor = OpenAIVisionProcessor()
+    source = Path(testdir, 'data/template_conversion_en_gp-document-IF10244.59.pdf.json')
+    with source.open("r") as fin:
+        jdoc = json.load(fin)
+        docs = [Document(**jdoc)]
+        for s in docs[0].sentences:
+            print(docs[0].text[s.start:s.end])
+            print("-------")
+        docs: List[Document] = processor.process(docs, parameters)
+        assert len(docs) == 1
+        print("======================")
+        doc0 = docs[0]
+        for s in doc0.sentences:
+            print(docs[0].text[s.start:s.end])
+            print("-------")
+        json_file = Path(testdir, 'data/template_conversion_en_gp-document-IF10244.59_gpt-4o.json')
+        with json_file.open("w") as fout:
+            print(docs[0].json(exclude_none=True, exclude_unset=True, indent=2), file=fout)
+    print("--- gpt-4o: %s seconds ---" % (time.time() - start_time))
+
+
 @pytest.mark.skip(reason="Not a test")
 def test_openai():
     converter = OpenAIVisionConverter()