data_pack.py

import os
import json
from PIL import Image
from datasets import Dataset, DatasetDict, Features, Sequence, Value
from datasets import Image as ImageData
from io import BytesIO  # 新增导入

from datasets import disable_caching
disable_caching()  # 完全禁用缓存

def generate_examples(json_path):
    with open(json_path, 'r', encoding='utf-8') as f:
        # i = 0
        for line in f:
            # i += 2 # renew cache
            # print(i) # verify cache renewed
            try:
                data = json.loads(line)
                # 转换图片路径并加载图像
                raw_path = data['images'][0]
                # ========== 新增图像处理部分开始 ==========
                with Image.open(raw_path, "r") as img:
                    # 统一转换为RGB模式
                    if img.mode != 'RGB':
                        img = img.convert('RGB')
                    
                    resized_img = img
                # ========== 新增图像处理部分结束 ==========
                
                    # 提取图片ID（文件名数字部分）
                    filename = os.path.basename(raw_path)  # 141126815887.jpg
                    img_id = os.path.splitext(filename)[0]  # 141126815887
                    yield {
                        "images": [resized_img],
                        "problem": data["query"],
                        "answer": data["response"],
                        "id": img_id
                    }
                # print(data["query"])
                # print(data["response"])
                # exit()
            except Exception as e:
                data = json.loads(line)
                print(data)
                print(f"Error processing {raw_path}: {str(e)}")
                continue

def main():
    train_json_path="./magic_mirror_data_train_r1.jsonl"
    test_json_path="./magic_mirror_data_test_r1.jsonl"

    base_dir = "./"
    train_parquet_path  =base_dir + "magic_mirror_data_train_r1.parquet"
    test_parquet_path   =base_dir + "magic_mirror_data_test_r1.parquet"
    
    train_json_path     = "./magic_mirror_data_train_resample_r1.jsonl"
    train_parquet_path  = "./magic_mirror_data_train_resample_r1.parquet"

    # 创建训练集和测试集
    test_ds     = Dataset.from_generator(generate_examples, gen_kwargs={"json_path": test_json_path})
    test_ds.cast_column("images", Sequence(ImageData())).to_parquet(test_parquet_path)
    
    train_ds    = Dataset.from_generator(generate_examples, gen_kwargs={"json_path": train_json_path})
    train_ds.cast_column("images", Sequence(ImageData())).to_parquet(train_parquet_path)


if __name__ == "__main__":
    main()