pyxtxt 0.3.2__tar.gz → 0.3.3__tar.gz

{pyxtxt-0.3.2/src/pyxtxt.egg-info → pyxtxt-0.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pyxtxt
-Version: 0.3.2
+Version: 0.3.3
 Summary: A Python library for extracting text from different types of files (PDF, DOCX, PPTX, XLSX, ODT, ecc.).
 Author-email: Giuseppe Levi <giuseppe.levi@gmail.com>
 License: MIT License

{pyxtxt-0.3.2 → pyxtxt-0.3.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "pyxtxt"
-version = "0.3.2"
+version = "0.3.3"
 description = "A Python library for extracting text from different types of files (PDF, DOCX, PPTX, XLSX, ODT, ecc.)."
 classifiers = [
   "Development Status :: 4 - Beta",

{pyxtxt-0.3.2 → pyxtxt-0.3.3}/src/pyxtxt/estrattori/ocr_ollama.py

@@ -18,7 +18,8 @@ OLLAMA_CONFIG = {
     'style': 'descriptive',  # descriptive, technical, simple, detailed
     'temperature': 0.1,  # Response creativity (0.0-1.0)
     'max_tokens': 1500,  # Maximum response length
-    'confidence_threshold': 0.7  # Minimum confidence for text extraction
+    'confidence_threshold': 0.7,  # Minimum confidence for text extraction
+    'context': 'general'  # Context hint: general, cookbook, document, diagram, etc.
 }
 
 def set_ollama_model(model_name: str):
@@ -48,13 +49,15 @@ def set_ollama_config(**kwargs):
     - language: Language hint ('auto', 'italian', 'english', 'spanish', etc.)
     - caption_length: Caption length ('short', 'medium', 'long')
     - style: Caption style ('descriptive', 'technical', 'simple', 'detailed')
+    - context: Content context ('general', 'cookbook', 'document', 'handwriting', 'technical')
     - temperature: Response creativity 0.0-1.0 (default: 0.1)
     - max_tokens: Maximum response length (default: 1500)
     - confidence_threshold: Text extraction confidence 0.0-1.0 (default: 0.7)
     
     Examples:
         set_ollama_config(language='italian', style='detailed')
-        set_ollama_config(caption_length='long', temperature=0.3)
+        set_ollama_config(context='document', caption_length='long')
+        set_ollama_config(context='handwriting', temperature=0.2)
     """
     global OLLAMA_CONFIG
     for key, value in kwargs.items():
@@ -77,7 +80,8 @@ def reset_ollama_config():
         'style': 'descriptive',
         'temperature': 0.1,
         'max_tokens': 1500,
-        'confidence_threshold': 0.7
+        'confidence_threshold': 0.7,
+        'context': 'general'
     }
     print("✅ Ollama configuration reset to defaults")
 
@@ -96,7 +100,11 @@ if ollama and Image:
             current_model = model or OLLAMA_MODEL
             
             # Convert buffer to PIL Image
-            image = Image.open(BytesIO(file_buffer.read()))
+            # Reset buffer position if it has read method
+            if hasattr(file_buffer, 'seek'):
+                file_buffer.seek(0)
+            image_data = file_buffer.read()
+            image = Image.open(BytesIO(image_data))
             
             # Convert to RGB if needed
             if image.mode != 'RGB':
@@ -117,14 +125,18 @@ if ollama and Image:
             
             # Different prompts based on mode
             if mode == "ocr":
-                prompt = f"""Extract ALL visible text from this image exactly as it appears. 
-Rules:
-- Only return text that is actually written/printed in the image
-- Preserve reading order (left to right, top to bottom)  
-- Maintain line breaks and formatting
-- Include numbers, symbols, special characters
-- Do NOT add descriptions, interpretations, or context
-- {lang_hint}If no text is visible, return 'NO_TEXT_FOUND'
+                prompt = f"""Look carefully at this image and extract ALL text that you can see, including:
+- Titles, headings, and main text content  
+- Small print, captions, labels, and annotations
+- Numbers, measurements, quantities, and symbols
+- Menu items, ingredient lists, cooking instructions
+- Any text in boxes, speech bubbles, or decorative elements
+
+IMPORTANT: 
+- Read carefully and include even small or partially visible text
+- Preserve the original formatting and line breaks where possible
+- {lang_hint}Process the text from left to right, top to bottom
+- If you cannot find any readable text at all, respond with 'NO_TEXT_FOUND'
 
 Extracted text:"""
             
@@ -146,12 +158,33 @@ Extracted text:"""
                 style_instruction = style_prompts.get(config['style'], style_prompts['descriptive'])
                 length_instruction = length_hints.get(config['caption_length'], length_hints['medium'])
                 
+                # Context-specific hints (only when explicitly set)
+                context_hint = ""
+                context = config.get('context', 'general').lower()
+                if context == 'cookbook' or context == 'recipe':
+                    context_hint = """
+   - If this appears to be a recipe/cookbook page, include: ingredients, cooking steps, quantities, cooking times
+   - Mention any photos of prepared dishes or cooking techniques shown
+   - Note any special formatting like ingredient lists, step numbers, or cooking tips"""
+                elif context == 'document':
+                    context_hint = """
+   - Focus on document structure: headers, paragraphs, sections, page numbers
+   - Note any official formatting, letterheads, signatures, or stamps"""
+                elif context == 'handwriting' or context == 'notes':
+                    context_hint = """
+   - Pay special attention to handwritten text which may be harder to read
+   - Note any sketches, diagrams, or informal formatting typical of personal notes"""
+                elif context == 'technical' or context == 'diagram':
+                    context_hint = """
+   - Focus on technical elements: labels, measurements, specifications, diagrams
+   - Include any mathematical formulas, technical symbols, or engineering notations"""
+
                 prompt = f"""Analyze this image and provide:
-1. All visible text exactly as written
+1. All visible text exactly as written (preserve formatting, line breaks, bullet points)
 2. Image description following these guidelines:
    - {style_instruction}
    - {length_instruction}
-   - {lang_hint}Focus on key visual elements, layout, and context
+   - {lang_hint}Focus on key visual elements, layout, and context{context_hint}
 
 Format:
 TEXT: [all visible text here, or NO_TEXT_FOUND if none]
@@ -183,13 +216,23 @@ DESCRIPTION: [image description following the guidelines above]"""
             return ""
     
     # Wrapper functions for each mode
-    def xtxt_image_ocr_only(file_buffer):
+    def xtxt_image_ocr_only(file_input):
         """Traditional OCR: extract only visible text using Ollama"""
-        return xtxt_image_ocr_ollama(file_buffer, mode="ocr")
+        # Handle both file paths and buffers
+        if isinstance(file_input, str):
+            with open(file_input, 'rb') as f:
+                return xtxt_image_ocr_ollama(f, mode="ocr")
+        else:
+            return xtxt_image_ocr_ollama(file_input, mode="ocr")
     
-    def xtxt_image_describe(file_buffer):
+    def xtxt_image_describe(file_input):
         """OCR + Description: text + image context using Ollama"""
-        return xtxt_image_ocr_ollama(file_buffer, mode="describe")
+        # Handle both file paths and buffers  
+        if isinstance(file_input, str):
+            with open(file_input, 'rb') as f:
+                return xtxt_image_ocr_ollama(f, mode="describe")
+        else:
+            return xtxt_image_ocr_ollama(file_input, mode="describe")
     
     # Register OCR-only version as default
     # Note: Will override traditional EasyOCR if both modules are loaded

{pyxtxt-0.3.2 → pyxtxt-0.3.3/src/pyxtxt.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pyxtxt
-Version: 0.3.2
+Version: 0.3.3
 Summary: A Python library for extracting text from different types of files (PDF, DOCX, PPTX, XLSX, ODT, ecc.).
 Author-email: Giuseppe Levi <giuseppe.levi@gmail.com>
 License: MIT License