metadata
license: mit
dataset_info:
features:
- name: image
dtype:
array3_d:
shape:
- 512
- 512
- 3
dtype: uint8
- name: filename
dtype: string
splits:
- name: train
num_bytes: 34195458528
num_examples: 18614
download_size: 6667979906
dataset_size: 34195458528
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
from datasets import load_dataset
from PIL import Image
import numpy as np
import os
from tqdm import tqdm
# 加载数据集
dataset_path = "path_to/style_fonts_img"
dataset = load_dataset(dataset_path)
# 创建保存目录
save_dir = os.path.join(dataset_path, "extracted_images")
os.makedirs(save_dir, exist_ok=True)
# 获取数据集大小
total_samples = len(dataset['train'])
print(f"数据集共有 {total_samples} 个样本")
# 使用tqdm创建进度条
for i, example in tqdm(enumerate(dataset['train']), total=total_samples, desc="处理图像"):
try:
# 获取图像数据
image_array = example['image']
# 转换为PIL图像
image = Image.fromarray(np.uint8(image_array))
# 获取文件名
filename = example['filename']
# 保存图像
image_path = os.path.join(save_dir, filename)
image.save(image_path)
# 保存文本
if 'text' in example and example['text']:
text_filename = os.path.splitext(filename)[0] + '.txt'
text_path = os.path.join(text_dir, text_filename)
with open(text_path, 'w', encoding='utf-8') as f:
f.write(example['text'])
# print(f"已保存 {filename}")
# 只处理前10个样本（可选）
if i >= 9:
break
except Exception as e:
print(f"处理样本 {i} 时出错: {e}")
print("处理完成！")