Upload 2 files

app.py

@@ -12,12 +12,6 @@ import easyocr
 from PIL import Image
 import numpy as np
 import re
-import matplotlib.pyplot as plt
-import matplotlib
-from io import BytesIO
-
-# Use non-interactive backend for matplotlib
-matplotlib.use('Agg')
 
 @st.cache_data(show_spinner=False)
 def get_pdf_text(pdf_docs):
@@ -47,24 +41,16 @@ def get_image_text(image_files):
         try:
             # Open image using PIL
             image = Image.open(image_file)
-            
-            # Preprocess image for better OCR (convert to grayscale and enhance contrast)
-            image = image.convert('L') # Grayscale
-            # Simple contrast enhancement using numpy
-            img_np = np.array(image)
-            # Normalize
-            img_np = (img_np - np.min(img_np)) / (np.max(img_np) - np.min(img_np)) * 255
-            img_np = img_np.astype(np.uint8)
-            
+            # Convert to numpy array for easyocr
+            image_np = np.array(image)
             # Extract text using EasyOCR with parameters optimized for math/handwriting
             results = reader.readtext(
-                img_np,
+                image_np,
                 detail=1,
                 paragraph=False,
                 contrast_ths=0.1,
                 adjust_contrast=0.5,
-                text_threshold=0.4, # Lower threshold to catch faint handwriting
-                low_text=0.3
+                text_threshold=0.5
             )
             # Combine all detected text with confidence filtering
             extracted_text = ' '.join([result[1] for result in results if result[2] > 0.3])
@@ -112,67 +98,6 @@ def get_conversation_chain(vectorstore, api_key):
     )
     return conversation_chain
 
-def render_math_as_image(text):
-    """
-    Detect LaTeX math in text and render complex equations as images.
-    Returns tuple: (has_math, processed_text_or_image)
-    """
-    # Check if text contains LaTeX math patterns
-    math_patterns = [
-        r'\$\$.*?\$\$', r'\$.*?\$', 
-        r'\\frac', r'\\sqrt', r'\\sum', r'\\int', 
-        r'\\begin\{.*?\}', r'\\\[.*?\\\]',
-        r'\[\s*\\begin'  # Catch [\begin cases
-    ]
-    has_complex_math = any(re.search(pattern, text, re.DOTALL) for pattern in math_patterns)
-    
-    if has_complex_math:
-        try:
-            # Clean the text for rendering
-            clean_text = text
-            
-            # Normalize delimiters
-            clean_text = clean_text.replace('$$', '$')
-            clean_text = clean_text.replace(r'\[', '$').replace(r'\]', '$')
-            clean_text = clean_text.replace(r'\(', '$').replace(r'\)', '$')
-            
-            # Fix specific issue seen in screenshot: [\begin -> $\begin
-            clean_text = re.sub(r'\[\s*\\begin', r'$\n\\begin', clean_text)
-            clean_text = re.sub(r'\\end\{.*?\}\s*\]', r'\\end{cases}$', clean_text) # Simplified fix for end
-            
-            # Remove spacing arguments like \\[2mm] -> \\
-            clean_text = re.sub(r'\\\\\[.*?\]', r'\\\\', clean_text)
-            
-            # Estimate height based on text length (rough approximation)
-            lines = clean_text.count('\n') + (len(clean_text) // 60)
-            fig_height = max(4, lines * 0.5)
-            
-            # Create figure
-            fig, ax = plt.subplots(figsize=(10, fig_height))
-            ax.axis('off')
-            
-            # Render text with math
-            ax.text(0.5, 0.5, clean_text, 
-                   fontsize=12, 
-                   ha='center', 
-                   va='center',
-                   wrap=True,
-                   usetex=False)  # Use matplotlib's math rendering
-            
-            # Save to bytes
-            buf = BytesIO()
-            plt.savefig(buf, format='png', bbox_inches='tight', dpi=200, facecolor='#f5f2e8')
-            buf.seek(0)
-            plt.close(fig)
-            
-            return True, buf
-        except Exception as e:
-            # If rendering fails, return text but try to clean it for markdown
-            print(f"Rendering failed: {e}")
-            return False, text
-    
-    return False, text
-
 def handle_userinput(user_question):
     # Display user message immediately
     st.chat_message("user").write(user_question)
@@ -181,15 +106,9 @@ def handle_userinput(user_question):
         response = st.session_state.conversation({'question': user_question})
         st.session_state.chat_history = response['chat_history']
     
-    # Display assistant response - render math as image if complex
+    # Display assistant response with LaTeX support
     answer = response['answer']
-    has_math, content = render_math_as_image(answer)
-    
-    with st.chat_message("assistant"):
-        if has_math:
-            st.image(content, use_column_width=True)
-        else:
-            st.markdown(content, unsafe_allow_html=True)
+    st.chat_message("assistant").markdown(answer, unsafe_allow_html=True)
 
 def main():
     load_dotenv()
@@ -306,12 +225,7 @@ def main():
             if i % 2 == 0:
                 st.chat_message("user").write(message.content)
             else:
-                has_math, content = render_math_as_image(message.content)
-                with st.chat_message("assistant"):
-                    if has_math:
-                        st.image(content, use_column_width=True)
-                    else:
-                        st.markdown(content, unsafe_allow_html=True)
+                st.chat_message("assistant").markdown(message.content, unsafe_allow_html=True)
 
     user_question = st.chat_input("Ask a question about your documents:")
     if user_question:

requirements.txt

@@ -12,5 +12,4 @@ python-dotenv
 tiktoken
 easyocr
 Pillow
-matplotlib