Add images

convert.py

@@ -0,0 +1,64 @@
+import os
+import pandas as pd
+from PIL import Image
+import codecs
+import numpy as np
+import glob
+
+def create_dataset():
+    print("Starting dataset creation...")
+    
+    # Create output directory for images
+    os.makedirs("processed_images", exist_ok=True)
+    
+    # Find all label files matching pattern
+    print("Finding label files...")
+    label_files = glob.glob("./original/test/round3?_best2019.label")
+    print(f"Found {len(label_files)} label files")
+    
+    # Parse image filenames and captions
+    data = []
+    for i, label_path in enumerate(label_files):
+        print(f"\nProcessing label file {i+1}/{len(label_files)}: {label_path}")
+        
+        # Read label file with Windows-874 encoding
+        print("Reading label file...")
+        with codecs.open(label_path, 'r', encoding='cp874') as f:
+            lines = f.readlines()
+        print(f"Found {len(lines)} entries")
+        
+        for j, line in enumerate(lines):
+            if j % 100 == 0:
+                print(f"Processing entry {j}/{len(lines)}")
+                
+            filename, caption = line.strip().split(' ', 1)
+            
+            # Get folder name from first 3 chars of filename
+            folder = filename[:3]
+            image_path = f"./original/test/{folder}/{filename}"
+            
+            # Load and verify image exists
+            if os.path.exists(image_path):
+                try:
+                    # Save image to processed directory and store path
+                    output_path = f"processed_images/{filename}"
+                    Image.open(image_path).save(output_path)
+                    data.append({
+                        'image_path': output_path,
+                        'text': caption,
+                        'label_file': os.path.basename(label_path)
+                    })
+                except Exception as e:
+                    print(f"Error processing image {image_path}: {e}")
+    
+    print(f"\nProcessed {len(data)} total images successfully")
+    
+    # Convert to dataframe and save as parquet
+    print("Converting to dataframe...")
+    df = pd.DataFrame(data)
+    print("Saving to parquet file...")
+    df.to_parquet("handwriting_dataset.parquet", index=False)
+    print("Dataset creation complete!")
+
+if __name__ == "__main__":
+    create_dataset()

meta.txt

@@ -0,0 +1,9 @@
+We want to create huggingface dataset for image handwritten recognition
+(1) Read the dataset from ./original/test/round31_best2019.label (It is encoded as Windows-784)
+
+round31_best2019.label
+r31-0001.png เป็นมนุษย์สุดประเสริฐเลิศคุณค่า กว่าบรรดาฝูงสัตว์เดรัจฉาน  
+r31-0002.png เป็นมนุษย์สุดประเสริฐเลิศคุณค่า กว่าบรรดาฝูงสัตว์เดรัจฉาน  
+
+(2) Load the image from folder that is the first 3 letters of filename e.g. r31-0001.png located at ./original/test/r31/r31-0001.png
+(3) Make image and caption pair save as the parquets