utils/csv/4D-BO.py

import csv
import json
from pathlib import Path

def csv_to_json(csv_path, json_path, file_stem):
    # 初始化JSON数据结构
    json_data = []
    
    with open(csv_path, 'r', encoding='utf-8') as csv_file:
        csv_reader = csv.DictReader(csv_file)
        
        for index, row in enumerate(csv_reader):
            # 构建media_path
            folder_name = row['folder_name']
            media_path = "./" + (Path("data") / file_stem / folder_name).as_posix()
            
            # 收集五个caption作为答案
            answer = [
                row['caption_1'],
                row['caption_2'],
                row['caption_3'],
                row['caption_4'],
                row['caption_5']
            ]
            
            # 构建JSON条目
            entry = {
                "index": index,
                "media_type": "Video",
                "media_paths": media_path,
                "description": "",
                "task_type": "Vision-Question-Answer",
                "question": ["Please generate descriptive captions for this multi-view video."],
                "question_type": "free-form",
                "annotations": {},
                "options": [],
                "answer": answer,
                "source": "4D-Bench",
                "domain": "Embodied_ai"
            }
            
            json_data.append(entry)
    
    # 写入JSON文件
    with open(json_path, 'w', encoding='utf-8') as json_file:
        json.dump(json_data, json_file, indent=2)

# 使用示例
if __name__ == "__main__":
    # 用户需要修改以下参数
    INPUT_CSV = "/mnt/data/users/zys/proj/vlm_reasoning/unprocessed_data/emb_ai/4d/4D_Object_Captioning/data/human_annotations.csv"      # 输入CSV文件路径
    OUTPUT_JSON = "/mnt/data/users/zys/proj/vlm_reasoning/dataset/4D_Object_Captioning.json"  # 输出JSON文件路径
    FILE_STEM = "4D_Object_Captioning"        # media_path中的file_stem部分
    
    csv_to_json(INPUT_CSV, OUTPUT_JSON, FILE_STEM)