convert_wang.py

import os
import pandas as pd
from PIL import Image
import codecs
import numpy as np
import glob
import io

def create_dataset():
    print("Starting dataset creation...")
    
    # Create output directory for images
    os.makedirs("processed_images", exist_ok=True)
    
    # Process wang dataset
    print("\nProcessing wang dataset...")
    wang_csv = "./original/wang/free_dataset.csv"
    if os.path.exists(wang_csv):
        df_wang = pd.read_csv(wang_csv, header=None)
        print(f"Found {len(df_wang)} entries in wang dataset")
        
        data = []
        file_count = 1
        
        for i, row in df_wang.iterrows():
            if i % 100 == 0:
                print(f"Processing wang entry {i+1}/{len(df_wang)}")
                
            _, text, _, filename = row
            image_path = os.path.join("/Users/kobkrit/git/iapp-dataset/thai_handwriting_dataset/original/wang/free_dataset", filename)
            
            if os.path.exists(image_path):
                try:
                    img = Image.open(image_path)
                    
                    # Convert image to PNG format
                    if img.format != 'PNG':
                        # Create a new RGB image with white background
                        png_img = Image.new('RGB', img.size, (255, 255, 255))
                        # Paste the original image onto the white background
                        png_img.paste(img, mask=img if img.mode=='RGBA' else None)
                        img = png_img
                        
                    img_byte_arr = io.BytesIO()
                    img.save(img_byte_arr, format='PNG')
                    img_bytes = {"bytes":bytearray(img_byte_arr.getvalue())}
                    
                    data.append({
                        'image': img_bytes,
                        'text': text,
                        'label_file': filename
                    })
                    
                    # print(data)
                    
                    # Save every 100 rows
                    if len(data) >= 100:
                        print(f"\nSaving batch {file_count} with {len(data)} images")
                        print("Converting to dataframe...")
                        df = pd.DataFrame(data)
                        print(f"Saving to parquet file train-{file_count:04d}.parquet...")
                        df.to_parquet(f"train-{file_count:04d}.parquet", index=False)
                        data = [] # Clear the data list
                        file_count += 1
                        
                except Exception as e:
                    print(f"Error processing wang image {image_path}: {e}")
        
        # Save any remaining data
        if len(data) > 0:
            print(f"\nSaving final batch {file_count} with {len(data)} images")
            print("Converting to dataframe...")
            df = pd.DataFrame(data)
            print(f"Saving to parquet file train-{file_count:04d}.parquet...")
            df.to_parquet(f"train-{file_count:04d}.parquet", index=False)
            
    print("Dataset creation complete!")

if __name__ == "__main__":
    create_dataset()