data_creator.py

import os
import string
import csv
import pandas as pd
import requests
import io
from zipfile import ZipFile
import json
import shutil
import uuid
from tqdm import tqdm
from fontTools.ttLib import TTFont
from fontTools.unicode import Unicode
from fontpreview import FontPreview
import random
import sys

def has_glyph(font, glyph):
    for table in font['cmap'].tables:
        if ord(glyph) in table.cmap.keys():
            return True
    return False

#For a given font file, create the alphabet and the numbers 0-9
def create_alphabet(font_file, image_folder):
    font = FontPreview(font_file)
    ttf_font = TTFont(font_file)
    font_name = font.font.getname()[0]
    included_chars = []
    for char in string.ascii_letters:
        if has_glyph(ttf_font, char):
            included_chars.append(char)
    for char in string.digits:
        if has_glyph(ttf_font, char):
            included_chars.append(char)
    split_folder = 'train'
    if len(included_chars) != 62:
        split_folder = 'test'
        
    save_path = os.path.join(image_folder, split_folder, font_name)
    if not os.path.exists(save_path):
        os.makedirs(save_path)
    for char in included_chars:
        if char in string.ascii_lowercase:
            image_file_name = 'lower_' + char + '.jpg'
        elif char in string.ascii_uppercase:
            image_file_name = 'upper_' + char + '.jpg'
        else:
            image_file_name = char + '.jpg'
        if save_path[-1] == ' ':
            save_path = save_path[:-1]
        final_path = os.path.join(save_path, image_file_name)
        if not os.path.exists(final_path):
            font.font_text = char
            font.bg_color = (0, 0, 0)  # white BG
            font.dimension = (512, 512)  # Dimension consistent with the default resolution for diffusion models
            font.fg_color = (255, 255, 255)  # Letter color
            font.set_font_size(300)  # font size ~ 300 pixels
            font.set_text_position('center')  # center placement
            font.save(final_path)




def create_alphabet_for_each_ttf():
    TTF_DIR = os.path.join(os.path.abspath(os.getcwd()), 'ttf-files')
    IMG_DIR = os.path.join(os.path.abspath(os.getcwd()), 'font-images')
    if not os.path.exists(IMG_DIR):
        os.mkdir(IMG_DIR)
    fnames = os.listdir(TTF_DIR)

    for fname in tqdm(fnames):
        TTF_PATH = os.path.join(TTF_DIR, fname)
        create_alphabet(TTF_PATH, IMG_DIR)

    

#Uses pandas to read through the CSV from sheets without the need of constantly redownloading
def get_font_ttfs():
    # Read the CSV file into a Pandas DataFrame
    df = pd.read_csv('font_dataset.csv')
    # Create data folder if it does not exist
    if not os.path.exists('ttf-files'):
        os.makedirs('ttf-files')
    # Loop through each row of the DataFrame
    for index, row in tqdm(df.iterrows()):
        # Get the link and filename for the current row
        link = row['Link']
        filename = row['Filename']
        if os.path.exists(os.path.join('ttf-files', filename)):
            continue
            
        
        # Download the zip file from the link
        response = requests.get(link, stream=True)
        with open('temp.zip', 'wb') as temp_file:
            shutil.copyfileobj(response.raw, temp_file)
        del response
        # Unzip the downloaded file
        with ZipFile('temp.zip', 'r') as zip_file:
            zip_file.extract(filename)
            
        # Move the file to the data folder
        source_path = os.path.join(os.getcwd(), filename)
        dest_path = os.path.join(os.getcwd(), 'ttf-files', filename)
        shutil.move(source_path, dest_path)
        
        # Remove the temporary zip file
        os.remove('temp.zip')


#Create the jsonl file and training folder for the images
def create_dataset():
    FONT_IMAGE_PATH = os.path.join(os.getcwd(), 'font-images')
    assert os.path.exists(FONT_IMAGE_PATH)
    TTF_PATH = os.path.join(os.getcwd(), 'ttf-files')
    assert os.path.exists(TTF_PATH)
    CSV_PATH = os.path.join(os.getcwd(), 'font_dataset.csv')

    
    # Step 1: Initialize the json file
    # Step 2: Loop through the Dataframe, for each row the Filename column corresponds to the actual
    #         folder name in 'font_images'.
    # Step 3: For each image in the respective folder, copy it over to the training folder (renaming it) and add its entry
    #         to the jsonl file

    #Step 1
    # if not os.path.exists(training_data_path):
    #     os.makedirs(training_data_path)

    PROP_LIST = ['Weight', 'Corner Rounding', 'Serif', 'Width', 'Capitals', 'Dynamics']

    #Step 2
    df = pd.read_csv(CSV_PATH)
    train_dataset = []
    test_dataset = []
    for idx, row_data in df.iterrows():
        ttf_path = os.path.join(TTF_PATH, row_data['Filename'])
        font_img_dir = FontPreview(ttf_path).font.getname()[0]
        split_folder = 'train'
        font_img_dir_path = os.path.join(FONT_IMAGE_PATH, split_folder, font_img_dir)
        font_img_dir_path = font_img_dir_path.strip()
        if not os.path.exists(font_img_dir_path):
            split_folder = 'test'
            font_img_dir_path = os.path.join(FONT_IMAGE_PATH, split_folder, font_img_dir)
        font_img_paths = [os.path.join(font_img_dir_path, fname) for fname in os.listdir(font_img_dir_path)]
        font_img_paths.sort()
        if sys.platform == 'win32':
            included_chars = [cur_img_path.split('\\')[-1].split('.')[0] for cur_img_path in font_img_paths]
        else:
            included_chars = [cur_img_path.split('/')[-1].split('.')[0] for cur_img_path in font_img_paths]
        font_rows = []
        for img_path, char in zip(font_img_paths, included_chars):
            for key in PROP_LIST:
                json_data_row = {
                    'uniqueId': str(uuid.uuid4()),
                    'image': img_path,
                    'ttf_path': ttf_path,
                    'font_characteristics': row_data['Descriptors'], 
                    'character': char,
                    'vit_label': str('upper_' + char.split('_')[1].upper()) if row_data['Capitals'] == 'all caps' and char.split('_')[0] == 'lower' else char,
                    'font_properties': row_data[key]
                }
            font_rows.append(json_data_row)
        if split_folder == 'train':
            train_dataset = train_dataset + font_rows
        else:
            test_dataset = test_dataset + font_rows
    #Create the jsonl file
    with open('train-metadata.jsonl', 'w') as f:
        for item in train_dataset:
            f.write(json.dumps(item) + '\n')
    with open('test-metadata.jsonl', 'w') as f:
        for item in test_dataset:
            f.write(json.dumps(item) + '\n')