import json

# Load the original JSON file
with open('/home/yiyangai/stephenqs/datasets/mehrankazemi___re_mi/default/0.0.0/re_mi-test-02.json', 'r') as f:
    data = json.load(f)

# Function to convert the dataset
def convert_format(data):
    converted_data = []
    for item in data:
        new_item = {
            "id": item["id"],
            "image": None,
            "conversations": [
                {
                    "from": "human",
                    "content": []
                },
                {
                    "from": "gpt",
                    "content": [
                        {
                            "text": item["label"],
                            "image": None
                        }
                    ]
                }
            ]
        }

        # Add the main question with the first image
        new_item["conversations"][0]["content"].append({
            "text": item["question"],
            "image": item.get("image_1", None)
        })

        # If there are more images, add them with null text
        for i in range(2, 7):
            img_key = f"image_{i}"
            if item.get(img_key):
                new_item["conversations"][0]["content"].append({
                    "text": None,
                    "image": item[img_key]
                })

        converted_data.append(new_item)
    
    return converted_data

# Convert the dataset
converted_data = convert_format(data)

# Save the converted data to a new JSON file
with open('converted_dataset-02.json', 'w') as f:
    json.dump(converted_data, f, indent=4)