Datasets:
Add convert_wang.py
Browse files- convert_wang.py +80 -0
convert_wang.py
ADDED
|
@@ -0,0 +1,80 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import pandas as pd
|
| 3 |
+
from PIL import Image
|
| 4 |
+
import codecs
|
| 5 |
+
import numpy as np
|
| 6 |
+
import glob
|
| 7 |
+
import io
|
| 8 |
+
|
| 9 |
+
def create_dataset():
|
| 10 |
+
print("Starting dataset creation...")
|
| 11 |
+
|
| 12 |
+
# Create output directory for images
|
| 13 |
+
os.makedirs("processed_images", exist_ok=True)
|
| 14 |
+
|
| 15 |
+
# Process wang dataset
|
| 16 |
+
print("\nProcessing wang dataset...")
|
| 17 |
+
wang_csv = "./original/wang/free_dataset.csv"
|
| 18 |
+
if os.path.exists(wang_csv):
|
| 19 |
+
df_wang = pd.read_csv(wang_csv, header=None)
|
| 20 |
+
print(f"Found {len(df_wang)} entries in wang dataset")
|
| 21 |
+
|
| 22 |
+
data = []
|
| 23 |
+
file_count = 1
|
| 24 |
+
|
| 25 |
+
for i, row in df_wang.iterrows():
|
| 26 |
+
if i % 100 == 0:
|
| 27 |
+
print(f"Processing wang entry {i+1}/{len(df_wang)}")
|
| 28 |
+
|
| 29 |
+
_, text, _, filename = row
|
| 30 |
+
image_path = os.path.join("/Users/kobkrit/git/iapp-dataset/thai_handwriting_dataset/original/wang/free_dataset", filename)
|
| 31 |
+
|
| 32 |
+
if os.path.exists(image_path):
|
| 33 |
+
try:
|
| 34 |
+
img = Image.open(image_path)
|
| 35 |
+
|
| 36 |
+
# Convert image to PNG format
|
| 37 |
+
if img.format != 'PNG':
|
| 38 |
+
# Create a new RGB image with white background
|
| 39 |
+
png_img = Image.new('RGB', img.size, (255, 255, 255))
|
| 40 |
+
# Paste the original image onto the white background
|
| 41 |
+
png_img.paste(img, mask=img if img.mode=='RGBA' else None)
|
| 42 |
+
img = png_img
|
| 43 |
+
|
| 44 |
+
img_byte_arr = io.BytesIO()
|
| 45 |
+
img.save(img_byte_arr, format='PNG')
|
| 46 |
+
img_bytes = {"bytes":bytearray(img_byte_arr.getvalue())}
|
| 47 |
+
|
| 48 |
+
data.append({
|
| 49 |
+
'image': img_bytes,
|
| 50 |
+
'text': text,
|
| 51 |
+
'label_file': filename
|
| 52 |
+
})
|
| 53 |
+
|
| 54 |
+
# print(data)
|
| 55 |
+
|
| 56 |
+
# Save every 100 rows
|
| 57 |
+
if len(data) >= 100:
|
| 58 |
+
print(f"\nSaving batch {file_count} with {len(data)} images")
|
| 59 |
+
print("Converting to dataframe...")
|
| 60 |
+
df = pd.DataFrame(data)
|
| 61 |
+
print(f"Saving to parquet file train-{file_count:04d}.parquet...")
|
| 62 |
+
df.to_parquet(f"train-{file_count:04d}.parquet", index=False)
|
| 63 |
+
data = [] # Clear the data list
|
| 64 |
+
file_count += 1
|
| 65 |
+
|
| 66 |
+
except Exception as e:
|
| 67 |
+
print(f"Error processing wang image {image_path}: {e}")
|
| 68 |
+
|
| 69 |
+
# Save any remaining data
|
| 70 |
+
if len(data) > 0:
|
| 71 |
+
print(f"\nSaving final batch {file_count} with {len(data)} images")
|
| 72 |
+
print("Converting to dataframe...")
|
| 73 |
+
df = pd.DataFrame(data)
|
| 74 |
+
print(f"Saving to parquet file train-{file_count:04d}.parquet...")
|
| 75 |
+
df.to_parquet(f"train-{file_count:04d}.parquet", index=False)
|
| 76 |
+
|
| 77 |
+
print("Dataset creation complete!")
|
| 78 |
+
|
| 79 |
+
if __name__ == "__main__":
|
| 80 |
+
create_dataset()
|