Datasets:
Tasks:
Image-Text-to-Text
Modalities:
Image
Formats:
imagefolder
Languages:
English
Size:
1K - 10K
ArXiv:
License:
File size: 5,646 Bytes
1e9a83f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 |
import os
import json
import random
from PIL import Image, ImageDraw, ImageFont
# 设置输出目录
OUTPUT_DIR = "palindrome_dataset/img"
os.makedirs(OUTPUT_DIR, exist_ok=True)
DIGIT_COLORS = ["black", "purple", "pink", "blue", "green"]
FONT_SIZE = 50
# 适配 Windows/Linux 字体路径
if os.name == "nt": # Windows
FONT_PATH = "C:/Windows/Fonts/arial.ttf"
else: # Linux
FONT_PATH = "/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf"
def generate_palindrome(length):
""" 生成一个回文数字串 """
half = [random.randint(1, 9) for _ in range(length // 2)]
palindrome = half + half[::-1] if length % 2 == 0 else half + [random.randint(1, 9)] + half[::-1]
return palindrome
def create_palindrome_image(file_name, palindrome):
""" 创建回文数字图片，最后一个数字变成问号 """
img_width = len(palindrome) * 60
img_height = 100
image = Image.new("RGB", (img_width, img_height), "white")
draw = ImageDraw.Draw(image)
try:
font = ImageFont.truetype(FONT_PATH, FONT_SIZE)
except OSError:
print("⚠️ 字体文件未找到，使用 PIL 默认字体")
font = ImageFont.load_default()
for i, num in enumerate(palindrome[:-1]):
color = random.choice(DIGIT_COLORS)
draw.text((i * 60 + 10, 25), str(num), font=font, fill=color)
# 最后一个数字换成问号
draw.text(((len(palindrome) - 1) * 60 + 10, 25), "?", font=font, fill="red")
image_path = f"{OUTPUT_DIR}/{file_name}"
image.save(image_path)
return image_path
def generate_json(file_name, data):
""" 生成 JSON 文件 """
with open(file_name, "w", encoding="utf-8") as f:
json.dump(data, f, ensure_ascii=False, indent=4)
def main():
# 生成 support.json（32 个完整回文）
support_data = []
for i in range(32):
length = random.randint(5, 9)
palindrome = generate_palindrome(length)
file_name = f"support_{i + 1}.png"
image_path = create_palindrome_image(file_name, palindrome)
support_data.append({
"id": f"support_{i + 1}",
"image": [image_path],
"question": "What is the missing number in this palindrome?",
"answer": palindrome[-1]
})
generate_json("support.json", support_data)
# 生成 query.json（100 个回文，独立）
query_data = []
for i in range(100):
length = random.randint(5, 9)
palindrome = generate_palindrome(length)
file_name = f"query_{i + 1}.png"
image_path = create_palindrome_image(file_name, palindrome)
query_data.append({
"id": f"query_{i + 1}",
"image": [image_path],
"question": "What is the missing number in this palindrome?",
"answer": palindrome[-1]
})
generate_json("query.json", query_data)
# 生成 training_shots_pool.json（独立的 32 组 shots，不和 support.json 重复）
training_shots_pool = []
for i in range(32):
length = random.randint(5, 9)
palindrome = generate_palindrome(length)
file_name = f"training_shot_{i + 1}.png"
image_path = create_palindrome_image(file_name, palindrome)
training_shots_pool.append({
"id": f"training_shot_{i + 1}",
"image": [image_path],
"question": "What is the missing number in this palindrome?",
"answer": palindrome[-1]
})
generate_json("training_shots_pool.json", training_shots_pool)
# 生成 training_query.json（独立的 50 个 query，不和 query.json 重复）
training_queries = []
for i in range(50):
length = random.randint(5, 9)
palindrome = generate_palindrome(length)
file_name = f"training_query_{i + 1}.png"
image_path = create_palindrome_image(file_name, palindrome)
training_queries.append({
"id": f"training_query_{i + 1}",
"image": [image_path],
"question": "What is the missing number in this palindrome?",
"answer": palindrome[-1]
})
generate_json("training_query.json", training_queries)
# 生成 training_data.json（50 组，每个 4 shots + 1 query）
training_data = []
for query in training_queries:
chosen_shots = random.sample(training_shots_pool, 4) # ✅ shots 现在只从 training_shots_pool 选
messages = [{"role": "user", "content": [{"type": "text", "text": "Learn from the demos and give only the answer to the final question."}]}]
for shot in chosen_shots:
messages[0]["content"].append({"type": "image", "image": shot["image"][0]})
messages[0]["content"].append({"type": "text", "text": f"Question: What is the missing number in this palindrome? Answer: {shot['answer']}\n"})
messages[0]["content"].append({"type": "image", "image": query["image"][0]})
messages[0]["content"].append({"type": "text", "text": "Question: What is the missing number in this palindrome? Answer: \n"})
messages.append({"role": "assistant", "content": [{"type": "text", "text": str(query['answer'])}]})
training_data.append({"id": query["id"], "messages": messages})
generate_json("training_data.json", training_data)
print("✅ 数据集生成完成！已保存 query.json、support.json、training_shots_pool.json、training_query.json 和 training_data.json！")
if __name__ == "__main__":
main()
|