TASK:lookbook上传

app/service/lookbooks/service.py

@@ -1,11 +1,17 @@
 import json
 import os
+import logging
+import tqdm
+import aiofiles
 
 from openai import OpenAI
-
 from app.service.lookbooks.utils.image_utils import base64_encode_image, generate_text_id
 from app.service.lookbooks.utils.openai_utils import wait_for_job_completion
 
+# 设置日志
+logger = logging.getLogger()
+
+# OpenAI 配置
 OPENAI_API_KEY = "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"
 OPENAI_API_BASE = "https://pangkaichen-openai-prox-98.deno.dev/v1"
 client = OpenAI(
@@ -84,7 +90,8 @@ def create_image_batch_requests(
         }
         tasks.append(task)
         id2img[current_id] = image_filename
-    print(f"In total {len(tasks)} images")
+    logger.info(f"In total {len(tasks)} images to process")
+
     if tasks:
         batch_file_name = os.path.join(output_path, "image_batch_requests.jsonl")
         with open(batch_file_name, 'w', encoding='utf-8') as file:
@@ -120,9 +127,77 @@ def create_image_batch_requests(
                             })
                             f.write(output + '\n')
                         except json.JSONDecodeError as error:
-                            print(f"Error parsing: {error} -- at line: {line}")
+                            logger.error(f"Error parsing: {error} -- at line: {line}")
         else:
-            print("Job failed")
+            logger.error("Job failed")
         return os.path.join(output_path, "image_description_results.jsonl")
     else:
         return None
+
+
+async def process_lookbook_task(lookbook_list, tag, year):
+    """后台异步任务，用于处理 lookbook 并保存到向量数据库"""
+    image_list = []
+    try:
+        for look_book_path in tqdm.tqdm(lookbook_list):
+            lookbook_name = os.path.splitext(os.path.basename(look_book_path))[0]
+            output_dir = os.path.join("fashion_documents/lookbook/images", lookbook_name)
+            os.makedirs(output_dir, exist_ok=True)
+            if not os.listdir(output_dir):
+                from unstructured.partition.pdf import partition_pdf
+                partition_pdf(
+                    filename=look_book_path,
+                    extract_images_in_pdf=True,
+                    infer_table_structure=False,
+                    chunking_strategy="by_title",
+                    max_characters=4000,
+                    new_after_n_chars=3800,
+                    combine_text_under_n_chars=2000,
+                    extract_image_block_output_dir=output_dir,
+                )
+            else:
+                current_images = os.listdir(output_dir)
+                image_list.extend([os.path.join(output_dir, x) for x in current_images])
+
+        # 1. 处理图片并生成批量请求
+        image_description_results_file = create_image_batch_requests(image_list, "fashion_documents/lookbook/results")
+
+        # 2. 保存结果到向量数据库
+        if image_description_results_file:
+            save_to_vector_db(image_description_results_file, tag, year)
+
+    except Exception as e:
+        logger.error(f"Error processing lookbooks: {str(e)}")
+        raise e
+
+
+def save_to_vector_db(image_description_results_file, tag, year):
+    """保存图像描述到向量数据库"""
+    image_ids = set()
+    image_summaries = []
+    image_metadatas = []
+
+    try:
+        with open(image_description_results_file, "r", encoding="utf-8") as f:
+            for image_content in f:
+                image_content = json.loads(image_content)
+                # 确保ID不重复
+                if image_content["custom_id"] not in image_ids:
+                    image_ids.add(image_content["custom_id"])
+                    image_summaries.append(image_content["summary"])
+                    image_metadatas.append({
+                        "data_type": "image",
+                        "url": image_content["url"].replace("\\", "/"),
+                        "source": "mitu",
+                        "tag": tag,
+                        "year": year,
+                        "gender": "female"
+                    })
+
+        # 将图像的描述和元数据添加到向量数据库中
+        collection.add_texts(texts=image_summaries, metadatas=image_metadatas, ids=list(image_ids))
+        logger.info("Successfully saved data to vector database")
+
+    except Exception as e:
+        logger.error(f"Error saving to vector database: {e}")
+        raise e