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) | |