Update app.py

app.py

@@ -2,10 +2,9 @@ import gradio as gr
 from doctr.models import ocr_predictor
 import numpy as np
 
-# Load lightweight CPU model
 model = ocr_predictor(det_arch='db_resnet50', reco_arch='crnn_vgg16_bn', pretrained=True)
 
-def extract_aligned_table(image):
+def extract_perfect_table(image):
     if image is None:
         return "Please upload an image."
     
@@ -15,9 +14,9 @@ def extract_aligned_table(image):
     
     markdown_rows = []
     
-    # Define Column Boundaries (Adjusted for typical Accountancy Layouts)
-    # These are percentages of the page width (0.0 to 1.0)
-    col_bounds = [0.25, 0.65, 0.85] # Boundaries between Col 1|2, 2|3, 3|4
+    # Precise Column Boundaries (Adjusted for 4-column accountancy papers)
+    # Bucket 1: < 22% | Bucket 2: 22%-62% | Bucket 3: 62%-82% | Bucket 4: > 82%
+    col_bounds = [0.22, 0.62, 0.82] 
 
     for page in json_export['pages']:
         words_list = []
@@ -30,7 +29,6 @@ def extract_aligned_table(image):
         
         if not words_list: continue
             
-        # 1. Sort and Group into Rows
         words_list.sort(key=lambda w: w['y'])
         rows = []
         current_row = [words_list[0]]
@@ -42,49 +40,53 @@ def extract_aligned_table(image):
                 current_row = [words_list[i]]
         rows.append(current_row)
         
-        # 2. Assign words to specific Column "Buckets"
         for row in rows:
-            # Create 4 empty slots for [Acc Num, Acc Name, Debit, Credit]
             slots = ["", "", "", ""]
+            row.sort(key=lambda w: w['x']) # Ensure words in row are left-to-right
             
             for w in row:
                 x = w['x']
-                if x < col_bounds[0]:
-                    slots[0] += w['text'] + " "
+                text = w['text']
+                
+                # Logic to handle the "Debit Credit" header merging
+                if "Debit" in text and x > 0.55 and x < 0.75:
+                    slots[2] += text + " "
+                elif "Credit" in text and x > 0.75:
+                    slots[3] += text + " "
+                # General bucket logic for the rest of the rows
+                elif x < col_bounds[0]:
+                    slots[0] += text + " "
                 elif x < col_bounds[1]:
-                    slots[1] += w['text'] + " "
+                    slots[1] += text + " "
                 elif x < col_bounds[2]:
-                    slots[2] += w['text'] + " "
+                    slots[2] += text + " "
                 else:
-                    slots[3] += w['text'] + " "
+                    slots[3] += text + " "
             
-            # Clean up extra spaces and add to markdown
             clean_slots = [s.strip() for s in slots]
             markdown_rows.append("| " + " | ".join(clean_slots) + " |")
 
     return "\n".join(markdown_rows)
 
-# UI with JS-based Copy
 with gr.Blocks() as demo:
-    gr.Markdown("## 📑 Accountancy Aligned Table Extractor")
-    gr.Markdown("This version uses fixed columns to prevent Debit and Credit from merging.")
+    gr.Markdown("## 📑 Final Accountancy Table Extractor")
+    gr.Markdown("Fixed header alignment for Debit/Credit columns.")
     
     with gr.Row():
         with gr.Column():
-            img_in = gr.Image(type="pil", label="Upload Trial Balance")
-            btn = gr.Button("Extract with Columns", variant="primary")
+            img_in = gr.Image(type="pil")
+            btn = gr.Button("Extract Table", variant="primary")
         with gr.Column():
-            out = gr.Textbox(label="Result", lines=20, elem_id="out_box")
+            out = gr.Textbox(label="Copy-Paste Ready", lines=20, elem_id="out_box")
             copy_btn = gr.Button("📋 Copy Table")
-            
             copy_btn.click(None, None, None, js="""
                 () => {
                     const text = document.querySelector('#out_box textarea').value;
                     navigator.clipboard.writeText(text);
-                    alert('Table copied! Empty cells are now preserved.');
+                    alert('Table copied! Header columns are now separated.');
                 }
             """)
 
-    btn.click(extract_aligned_table, inputs=img_in, outputs=out)
+    btn.click(extract_perfect_table, inputs=img_in, outputs=out)
 
 demo.launch()