146 lines
5.2 KiB
Python
146 lines
5.2 KiB
Python
import json
|
||
import os
|
||
import logging
|
||
|
||
import requests
|
||
|
||
from Config.Config import GLM_API_KEY, GLM_MODEL_NAME, GLM_BASE_URL, OBS_PREFIX, OBS_AK, OBS_SK, OBS_SERVER, OBS_BUCKET
|
||
from Util.ObsUtil import ObsUploader
|
||
|
||
|
||
# 配置日志
|
||
def setup_logger():
|
||
logger = logging.getLogger(__name__)
|
||
if not logger.handlers:
|
||
logger.setLevel(logging.INFO)
|
||
handler = logging.StreamHandler()
|
||
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
|
||
handler.setFormatter(formatter)
|
||
logger.addHandler(handler)
|
||
return logger
|
||
|
||
|
||
logger = setup_logger()
|
||
|
||
|
||
# 批量处理图片
|
||
def batch_glm(output_dir, img_list):
|
||
qvq_result = f"{output_dir}/QvqResult.json"
|
||
if os.path.exists(qvq_result):
|
||
logger.info(f"GLM结果文件已存在，直接返回: {qvq_result}")
|
||
return qvq_result
|
||
|
||
img_url_list = []
|
||
for file_path in img_list:
|
||
# 创建上传器实例
|
||
uploader = ObsUploader(OBS_AK, OBS_SK, "https://" + OBS_SERVER)
|
||
# 上传参数
|
||
object_key = OBS_PREFIX + "/" + file_path
|
||
# 执行上传
|
||
success, result = uploader.upload_file(OBS_BUCKET, object_key, file_path)
|
||
# 处理结果
|
||
if success:
|
||
logger.info(f'{file_path}上传成功!')
|
||
# 获取上传文件的 URL
|
||
file_url = f"https://{OBS_BUCKET}.{OBS_SERVER}/{object_key}"
|
||
img_url_list.append(file_url)
|
||
else:
|
||
logger.error(f'{file_path}上传失败!')
|
||
if 'errorCode' in result:
|
||
logger.info(f'错误代码: {result["errorCode"]}')
|
||
logger.info(f'错误信息: {result["errorMessage"]}')
|
||
else:
|
||
logger.error(f'错误信息: {result["error"]}')
|
||
# 多张图片开始解析
|
||
answer_content = ""
|
||
for img_url in img_url_list:
|
||
answer_content = answer_content + glm_single(img_url)
|
||
|
||
# 保存结果到JSON文件
|
||
with open(qvq_result, "w", encoding='utf-8') as f:
|
||
f.write(answer_content)
|
||
return qvq_result
|
||
|
||
|
||
def glm_single(img_url):
|
||
logger.info(f"开始调用GLM API识别几何图形: {img_url}")
|
||
url = GLM_BASE_URL
|
||
headers = {
|
||
"Authorization": "Bearer " + GLM_API_KEY,
|
||
"Content-Type": "application/json"
|
||
}
|
||
prompt = """
|
||
### 几何图形识别专家指令（输入：纯几何图形照片）
|
||
|
||
**任务目标**
|
||
精确提取图形中的几何元素及其空间关系，为GeoGebra重建建立数学模型
|
||
|
||
## 一、坐标系建立规则（必须遵守）
|
||
1. 原点设定：
|
||
- 若存在明显顶点，选最左下角的点为原点(0,0)
|
||
- 若图形对称，选对称中心为原点
|
||
- 示例：原点O = 三角形ABC的顶点A
|
||
|
||
2. 坐标轴定向：
|
||
- 优先顺序：水平线段 > 垂直线段 > 最长线段
|
||
- 具体规则：
|
||
if 存在水平线段: 以该线段为x轴正方向
|
||
elif 存在垂直结构: 以最左侧垂直线为y轴
|
||
else: 以最长线段为基准轴
|
||
|
||
## 二、元素列举
|
||
1. 按点，线，三角形，四边形，梯形，平行四边形，矩形，正方形，圆等由简单到复杂的顺序列举所有图形
|
||
2. 详细描述元素之间的关系，比如点D在线段AB上
|
||
3. 详细描述元素之间的位置关系，比如D 在A点正上方，B在CD边的上方中间位置
|
||
|
||
"""
|
||
|
||
data = {
|
||
"model": GLM_MODEL_NAME,
|
||
"messages": [
|
||
{
|
||
"role": "user",
|
||
"content":
|
||
[
|
||
{
|
||
"type": "text",
|
||
"text": prompt
|
||
},
|
||
{
|
||
"type": "image_url",
|
||
"image_url": {
|
||
"url": img_url
|
||
}
|
||
}
|
||
]
|
||
}
|
||
],
|
||
"stream": True # 启用流式调用
|
||
}
|
||
|
||
answer_content = ''
|
||
try:
|
||
with requests.post(url, headers=headers, json=data, stream=True) as response:
|
||
for chunk in response.iter_lines():
|
||
if chunk:
|
||
decoded = chunk.decode('utf-8')
|
||
if decoded.startswith('[DONE]'):
|
||
logger.info("GLM API调用完成！")
|
||
break
|
||
try:
|
||
decoded = decoded[5:]
|
||
json_data = json.loads(decoded)
|
||
content = json_data["choices"][0]["delta"]['content']
|
||
if content and len(content) > 0:
|
||
print(content, end="")
|
||
answer_content = answer_content + content
|
||
except Exception as e:
|
||
logger.error(f"解析响应 chunk 时出错: {e}")
|
||
|
||
answer_content = answer_content.replace("```json", "")
|
||
answer_content = answer_content.replace("```", "")
|
||
return answer_content
|
||
except Exception as e:
|
||
logger.error(f"GLM API调用失败: {e}")
|
||
raise
|