svjack/Genshin_Impact_Scaramouche_Images_Captioned

from datasets import load_dataset
from PIL import Image, ImageOps

def resize_and_pad(image, target_size=(1024, 1024)):
    # 计算原始图像的宽高比
    width, height = image.size
    target_width, target_height = target_size
    ratio = min(target_width / width, target_height / height)
    
    # 等比例缩放图像
    new_size = (int(width * ratio), int(height * ratio))
    resized_image = image.resize(new_size)
    
    # 创建一个新的黑色背景图像
    new_image = Image.new("RGB", target_size, (0, 0, 0))
    
    # 将缩放后的图像粘贴到新图像的中心
    new_image.paste(resized_image, ((target_width - new_size[0]) // 2, (target_height - new_size[1]) // 2))
    
    return new_image

# 加载数据集
ds = load_dataset("svjack/Genshin_Impact_Scaramouche_Images_Captioned")

# 对数据集中的 image 列进行处理
def process_example(example):
    example['image'] = resize_and_pad(example['image'])
    return example

# 应用处理函数到整个数据集
ds = ds.map(process_example)

ds = ds["train"]
import os
from uuid import uuid1
os.makedirs("Genshin_Impact_Scaramouche_Images_Captioned_Local")

for ele in ds:
  uuid_ = uuid1()
  im_name = os.path.join("Genshin_Impact_Scaramouche_Images_Captioned_Local", "{}.png".format(uuid_))
  txt_name = os.path.join("Genshin_Impact_Scaramouche_Images_Captioned_Local", "{}.txt".format(uuid_))
  ele["image"].save(im_name)
  with open(txt_name, "w") as f:
    f.write(ele["joy-caption"])