svjack/Genshin_Impact_Yumemizuki_Mizuki_Captioned_ZH_EN

import os
import cv2
import uuid
import numpy as np
from PIL import Image
from decord import VideoReader
from tqdm import tqdm

def save_video_with_uuid(video_reader, output_dir, uuid, frame_rate=30):
    """将 decord VideoReader 保存为 MP4 文件"""
    output_path = os.path.join(output_dir, f"{uuid}.mp4")
    
    # 获取视频属性
    width = video_reader[0].shape[1]
    height = video_reader[0].shape[0]
    
    # 初始化 VideoWriter
    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
    video_writer = cv2.VideoWriter(output_path, fourcc, frame_rate, (width, height))
    
    # 逐帧写入
    for frame in video_reader:
        frame = frame.asnumpy()  # 转换为 numpy 数组
        frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)  # 转换颜色空间
        video_writer.write(frame)
    
    video_writer.release()

def save_image_with_uuid(image, output_dir, uuid):
    """将 PIL Image 保存为 PNG 文件"""
    output_path = os.path.join(output_dir, f"{uuid}.png")
    image.save(output_path)

def save_text_with_uuid(text, output_dir, uuid):
    """将文本保存为 TXT 文件"""
    output_path = os.path.join(output_dir, f"{uuid}.txt")
    with open(output_path, "w") as f:
        f.write("In the style of  Yumemizuki Mizuki, " + text)

def process_dataset(dataset, output_dir, data_type="video"):
    """处理数据集并保存文件"""
    # 创建输出目录
    os.makedirs(output_dir, exist_ok=True)
    
    # 遍历数据集
    for example in tqdm(dataset, desc=f"Processing {data_type} dataset"):
        # 生成唯一的 UUID
        unique_id = str(uuid.uuid4())
        
        # 保存数据文件
        if data_type == "video":
            video_reader = example["video"]
            save_video_with_uuid(video_reader, output_dir, unique_id)
        elif data_type == "image":
            image = example["image"]
            save_image_with_uuid(image, output_dir, unique_id)
        
        # 保存描述文件
        save_text_with_uuid(example["en_description"], output_dir, unique_id)

# 加载数据集
image_ds = load_dataset("svjack/Genshin_Impact_Yumemizuki_Mizuki_Captioned_ZH_EN", split="train")
video_ds = load_dataset("svjack/Genshin_Impact_Yumemizuki_Mizuki_Video_Captioned_ZH_EN", split="train")

# 处理视频数据集
process_dataset(video_ds, "./output_videos", data_type="video")

# 处理图像数据集
process_dataset(image_ds, "./output_images", data_type="image")