Merge branch 'develop'

# Conflicts:
#	app/service/design_fast/utils/synthesis_item.py
#	app/service/prompt_generation/chatgpt_for_translation.py

.dockerignore

@@ -1,2 +1,6 @@
 seg_cache
 test
+.venv
+__pycache__/
+*.pyc
+.git/

.gitea/workflows/develop_build_manual.yaml

@@ -7,7 +7,7 @@ jobs:
     runs-on: ubuntu-latest
 
     env:
-      REMOTE_DEPLOY_PATH: /workspace/Trinity/Fastapi_AiDA_Trinity_Dev
+      REMOTE_DEPLOY_PATH: /workspace/AiDA_Workspace/Python_Server_Workspace/Dev
 
     steps:
       - name: 1.检出代码
@@ -35,6 +35,4 @@ jobs:
             cd ${{ env.REMOTE_DEPLOY_PATH }}
             
             docker-compose down 2>&1
-            docker-compose up -d --build --remove-orphans 2>&1
+            docker-compose up -d  2>&1
-            
-            docker image prune -f 2>&1

.gitea/workflows/develop_build_scheduled.yaml

@@ -1,15 +1,15 @@
 name: 定时 AiDA python develop 分支构建部署
 on:
   # 使用 schedule 触发器，遵循标准的 Cron 格式 (分钟 小时-8 日期 月份 星期)
-  schedule:
+  # schedule:
-    - cron: '30 9 * * *'
+  #   - cron: '30 9 * * *'
 
 jobs:
   scheduled_deploy:
     runs-on: ubuntu-latest
 
     env:
-      REMOTE_DEPLOY_PATH: /workspace/Trinity/Fastapi_AiDA_Trinity_Dev
+      REMOTE_DEPLOY_PATH: /workspace/AiDA_Workspace/Python_Server_Workspace/Dev
 
     steps:
       - name: 1.检出代码

.gitea/workflows/ltx_develop_build_manual.yaml

@@ -1,23 +1,19 @@
-name: git commit AiDA python develop 分支构建部署
+name: 手动 AiDA python develop 分支构建部署
 on:
   workflow_dispatch:
-  push:
-    branches:
-      - develop
 
 jobs:
   scheduled_deploy:
     runs-on: ubuntu-latest
-    if: "contains(github.event.head_commit.message, '[run build]')"
 
     env:
-      REMOTE_DEPLOY_PATH: /workspace/Trinity/Fastapi_AiDA_Trinity_Dev
+      REMOTE_DEPLOY_PATH: /workspace/AiDA_Workspace/Python_Server_Workspace/Dev
 
     steps:
       - name: 1.检出代码
         uses: actions/checkout@v4
         with:
-          ref: 'develop'
+          ref: 'dev-ltx'
 
       - name: 2.复制文件到服务器
         uses: appleboy/scp-action@v0.1.7
@@ -28,7 +24,7 @@ jobs:
           source: "."
           target: ${{ env.REMOTE_DEPLOY_PATH }}
       
-      - name: Restart Docker containers
+      - name: 3.重启docker-compose
         uses: appleboy/ssh-action@v0.1.10
         with:
           host: ${{ secrets.SERVER_HOST }}

.gitea/workflows/prod_build_manual.yaml

@@ -0,0 +1,40 @@
+name: 定时 AiDA python prod 分支构建部署
+on:
+  workflow_dispatch:
+
+jobs:
+  scheduled_deploy:
+    runs-on: ubuntu-latest
+
+    env:
+      REMOTE_DEPLOY_PATH: /workspace/AiDA_Workspace/Python_Server_Workspace/Prod
+
+    steps:
+      - name: 1.检出代码
+        uses: actions/checkout@v4
+        with:
+          ref: 'master'
+
+      - name: 2.复制文件到服务器
+        uses: appleboy/scp-action@v0.1.7
+        with:
+          host: ${{ secrets.SERVER_HOST }}
+          username: ${{ secrets.SERVER_USER }}
+          password: ${{ secrets.SERVER_PASSWORD }}
+          source: "."
+          target: ${{ env.REMOTE_DEPLOY_PATH }}
+      
+      - name: Restart Docker containers
+        uses: appleboy/ssh-action@v0.1.10
+        with:
+          host: ${{ secrets.SERVER_HOST }}
+          username: ${{ secrets.SERVER_USER }}
+          password: ${{ secrets.SERVER_PASSWORD }}
+          script: |
+            # 进入项目目录
+            cd ${{ env.REMOTE_DEPLOY_PATH }}
+            
+            docker-compose down 2>&1
+            docker-compose up -d 2>&1
+            
+            docker image prune -f 2>&1

.gitea/workflows/prod_build_scheduled.yaml

@@ -0,0 +1,42 @@
+name: 定时 AiDA python prod 分支构建部署
+on:
+  # 使用 schedule 触发器，遵循标准的 Cron 格式 (分钟 小时-8 日期 月份 星期)
+  schedule:
+    - cron: '07 13 23 1 *'
+
+jobs:
+  scheduled_deploy:
+    runs-on: ubuntu-latest
+
+    env:
+      REMOTE_DEPLOY_PATH: /workspace/AiDA_Workspace/Python_Server_Workspace/Prod
+
+    steps:
+      - name: 1.检出代码
+        uses: actions/checkout@v4
+        with:
+          ref: 'master'
+
+      - name: 2.复制文件到服务器
+        uses: appleboy/scp-action@v0.1.7
+        with:
+          host: ${{ secrets.SERVER_HOST }}
+          username: ${{ secrets.SERVER_USER }}
+          password: ${{ secrets.SERVER_PASSWORD }}
+          source: "."
+          target: ${{ env.REMOTE_DEPLOY_PATH }}
+      
+      - name: Restart Docker containers
+        uses: appleboy/ssh-action@v0.1.10
+        with:
+          host: ${{ secrets.SERVER_HOST }}
+          username: ${{ secrets.SERVER_USER }}
+          password: ${{ secrets.SERVER_PASSWORD }}
+          script: |
+            # 进入项目目录
+            cd ${{ env.REMOTE_DEPLOY_PATH }}
+            
+            docker-compose down 2>&1
+            docker-compose up -d 2>&1
+            
+            docker image prune -f 2>&1

README.md

@@ -20,7 +20,6 @@
          $ conda activate trinity_client_aida 
          $ pip install -r requirements.txt
          $ conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia -y
-         $ pip install mmcv==1.4.2 -f https://download.openmmlab.com/mmcv/dist/cu117/torch1.13/index.html
 
 
 1. 启动服务器

app/api/api_design.py

@@ -4,6 +4,7 @@ import logging
 import requests
 from fastapi import APIRouter, HTTPException, BackgroundTasks
 
+from app.core.config import settings
 from app.schemas.design import DesignModel, ModelProgressModel, DesignStreamModel, SAMRequestModel
 from app.schemas.response_template import ResponseModel
 from app.service.design_fast.design_generate import design_generate, design_generate_v2
@@ -394,7 +395,8 @@ async def seg_anything(request_data: SAMRequestModel):
     通过传入图片路径和点击的点坐标，返回分割后的掩码数据。
 
     ### 参数说明:
-    - **user_id**:用户id 用于存储分割图
+    - **bucket**: minio bucket name
+    - **object_name**: minio object name
     - **image_path**: 图片在服务器或云端的相对路径。
     - **type**: 推理类型
     - **box**: 框选矩形点位信息
@@ -407,7 +409,8 @@ async def seg_anything(request_data: SAMRequestModel):
     ```json
     point
     {
-        "user_id": 1,
+        "bucket": "test",
+        "object_name": "7068-400a-ac94-c01647fa5f6f.png",
         "image_path": "aida-users/89/sketch/4e8fe37d-7068-400a-ac94-c01647fa5f6f.png",
         "type":"point",
         "points": [[310, 403], [493, 375], [261, 266], [404, 484]],
@@ -416,7 +419,8 @@ async def seg_anything(request_data: SAMRequestModel):
 
     box
     {
-        "user_id": 1,
+        "bucket": "test",
+        "object_name": "7068-400a-ac94-c01647fa5f6f.png",
         "image_path": "aida-users/89/sketch/4e8fe37d-7068-400a-ac94-c01647fa5f6f.png",
         "type":"box",
         "box": [350, 286, 544, 520]
@@ -425,7 +429,7 @@ async def seg_anything(request_data: SAMRequestModel):
     """
     try:
         logger.info(f"seg_anything request item is : @@@@@@:{json.dumps(request_data.dict(), indent=4)}")
-        data = requests.post("http://10.1.1.240:10075/predict", json=request_data.dict())
+        data = requests.post(f"http://{settings.B_4_X_4090_SERVICE_HOST}:10075/predict", json=request_data.dict())
         logger.info(f"seg_anything response @@@@@@:{json.dumps(json.loads(data.content), indent=4)}")
         return ResponseModel(data=json.loads(data.content))
     except Exception as e:

app/api/api_generate_image.py

@@ -1,9 +1,12 @@
 import json
 import logging
 
+import httpx
+import requests
 from fastapi import APIRouter, BackgroundTasks, HTTPException
 
-from app.schemas.generate_image import GenerateImageModel, GenerateProductImageModel, GenerateSingleLogoImageModel, GenerateRelightImageModel, GenerateMultiViewModel, BatchGenerateProductImageModel, BatchGenerateRelightImageModel, AgentTollGenerateImageModel
+from app.core.config import settings
+from app.schemas.generate_image import GenerateImageModel, GenerateProductImageModel, GenerateSingleLogoImageModel, GenerateRelightImageModel, GenerateMultiViewModel, BatchGenerateProductImageModel, BatchGenerateRelightImageModel, AgentTollGenerateImageModel, Flux2ToProductImgModel, GenerateSloganImageModel, GenerateImageFlux2KleinModel
 from app.schemas.pose_transform import BatchPoseTransformModel
 from app.schemas.response_template import ResponseModel
 from app.service.generate_batch_image.service import start_product_batch_generate, start_relight_batch_generate, start_pose_transform_batch_generate
@@ -20,6 +23,61 @@ logger = logging.getLogger()
 '''generate image'''
 
 
+# flux2 klein
+@router.post("/generate_image_flux2_klein")
+async def generate_image_flux2_klein(request_item: GenerateImageFlux2KleinModel):
+    """
+        创建一个具有以下参数的请求体:
+        - **bucket_name**: OSS桶名 (必填)
+        - **object_name**: OSS对象名（文件路径）(必填)
+
+        - **width**: 图片宽度，默认1024像素 (非必填,1024)
+        - **height**: 图片高度，默认1024像素 (非必填,默认1024)
+        - **prompt**: 文本提示词，用于模型推理等场景 (非必填,默认"")
+        - **steps**: 推理步数，控制模型生成过程的迭代次数 (非必填,默认4)
+        - **guidance**: 引导系数，调节提示词对生成结果的影响程度 (非必填,默认 4.0 )
+
+        ### 示例参数：
+        ```
+            {
+                "bucket_name": "aida-users",
+                "object_name": "89/moodboard/5fdc698c-cb9b-4b36-afa9ce4-1-89.png",
+                "prompt": "a single item of sketch of dress, 4k, white background"
+            }
+        ```
+        ### 输出示例:
+        ```
+            {
+                "code": 200,
+                "msg": "OK!",
+                "data": {
+                    "output_path": "aida-users/89/moodboard/5fdc698c-cb9b-4b36-afa9ce4-1-89.png"
+                }
+            }
+        ```
+    """
+    try:
+        logger.info(f"generate_image_flux2_gen_img request: {json.dumps(request_item.model_dump(), indent=4)}")
+        async with httpx.AsyncClient(timeout=120) as client:
+            resp = await client.post(
+                f"http://{settings.FLUX2_GEN_IMG_MODEL_URL}/predict",
+                json=request_item.model_dump(),
+            )
+        if resp.status_code == 200:
+            result = resp.json()
+            logger.info(f"flux2_gen_img response: {json.dumps(result, indent=4)}")
+            return ResponseModel(data=result)
+        else:
+            error = resp.json()
+            logger.info(f"flux2_gen_img response: {json.dumps(error, indent=4)}")
+            return ResponseModel(data=error, msg="ERROR!", code=500)
+
+    except Exception as e:
+        logger.warning(f"generate_image_flux2_gen_img Run Exception @@@@@@:{e}")
+        raise HTTPException(status_code=404, detail=str(e))
+
+
+# sdxl
 @router.post("/generate_image")
 def generate_image(request_item: GenerateImageModel, background_tasks: BackgroundTasks):
     """
@@ -154,6 +212,62 @@ def generate_single_logo_image(tasks_id: str):
     return ResponseModel(data=data['data'])
 
 
+"""slogan """
+
+
+@router.post("/generate_slogan")
+async def generate_slogan(request_data: GenerateSloganImageModel):
+    """
+        ### 请求体示例:
+        ```json
+        {
+          "num_point": 16,
+          "image_url": "aida-slogan/6886785f-0aac-4052-b6fd-7ae20a841d8d.png",
+          "prompt": "123",
+          "tasks_id": "string-89"
+        }
+        ```
+    """
+    try:
+        logger.info(f"generate_slogan request item is : @@@@@@:{json.dumps(request_data.dict(), indent=4)}")
+        data = requests.post(f"http://{settings.A6000_SERVICE_HOST}:10020/api/slogan", json=request_data.dict())
+        logger.info(f"generate_slogan response @@@@@@:{json.dumps(json.loads(data.content), indent=4)}")
+        return ResponseModel(data=json.loads(data.content))
+    except Exception as e:
+        logger.warning(f"generate_slogan Run Exception @@@@@@:{e}")
+
+
+"""product image flux2.0"""
+
+# @router.post("/img_to_product")
+# async def img_to_product(request_data: Flux2ToProductImgModel):
+#     """
+#         创建一个具有以下参数的请求体:
+#         - **tasks_id**: 任务id 用于取消生成任务和获取生成结果
+#         - **prompt**: 想要生成图片的描述词
+#         - **image_path**: 被生成图片的S3或minio url地址
+#         - **infer_step**: 推理步数
+#
+#         ### 请求体示例:
+#         ```json
+#         point
+#         {
+#             "prompt": "Create realistic studio photo with real people model standing and wearing this garment, in white studio, Keep original model if present, or generate appropriate model, Standing pose, facing camera.",
+#             "image_path":"aida-results/result_38151e0a-f83b-11f0-89f6-0242ac130002.png",
+#             "infer_step":4,
+#             "tasks_id":"123456-123"
+#         }
+#         ```
+#     """
+#     try:
+#         logger.info(f"img_to_product request item is : @@@@@@:{json.dumps(request_data.dict(), indent=4)}")
+#         data = requests.post(f"http://{settings.A6000_SERVICE_HOST}:10090/api/v1/to_product", json=request_data.dict())
+#         logger.info(f"img_to_product response @@@@@@:{json.dumps(json.loads(data.content), indent=4)}")
+#         return ResponseModel(data=json.loads(data.content))
+#     except Exception as e:
+#         logger.warning(f"img_to_product Run Exception @@@@@@:{e}")
+
+
 '''product image'''
 
 
@@ -178,7 +292,7 @@ def generate_product_image(request_item: GenerateProductImageModel, background_t
         }
     """
     try:
-        logger.info(f"generate_product_image request item is : @@@@@@:{json.dumps(request_item.dict(),indent=4)}")
+        logger.info(f"generate_product_image request item is : @@@@@@:{json.dumps(request_item.dict(), indent=4)}")
         service = GenerateProductImage(request_item)
         background_tasks.add_task(service.get_result)
     except Exception as e:

app/core/config.backup.py

@@ -1,235 +0,0 @@
-import os
-
-import pika
-from dotenv import load_dotenv
-from pydantic import BaseSettings
-
-BASE_DIR = os.path.abspath(os.path.join(os.path.dirname(__file__), '../../'))
-load_dotenv(os.path.join(BASE_DIR, '.env'))
-
-
-class Settings(BaseSettings):
-    PROJECT_NAME: str = 'FASTAPI BASE'
-    SECRET_KEY: str = ''
-    API_PREFIX: str = ''
-    BACKEND_CORS_ORIGINS: list[str] = ['*']
-    DATABASE_URL: str = ''
-    ACCESS_TOKEN_EXPIRE_SECONDS: int = 60 * 60 * 24 * 7  # Token expired after 7 days
-    SECURITY_ALGORITHM: str = 'HS256'
-    LOGGING_CONFIG_FILE: str = os.path.join(BASE_DIR, 'logging_env.py')
-
-
-OSS = "minio"
-DEBUG = False
-if DEBUG:
-    LOGS_PATH = "logs/"
-    CATEGORY_PATH = "service/attribute/config/descriptor/category/category_dis.csv"
-    SEG_CACHE_PATH = "../seg_cache/"
-    POSE_TRANSFORM_VIDEO_PATH = "../pose_transform_video/"
-    RECOMMEND_PATH_PREFIX = "service/recommend/"
-    CHROMADB_PATH = "./chromadb/"
-else:
-    LOGS_PATH = "app/logs/"
-    CATEGORY_PATH = "app/service/attribute/config/descriptor/category/category_dis.csv"
-    SEG_CACHE_PATH = "/seg_cache/"
-    POSE_TRANSFORM_VIDEO_PATH = "/pose_transform_video/"
-    RECOMMEND_PATH_PREFIX = "app/service/recommend/"
-    CHROMADB_PATH = "/chromadb/"
-
-# RABBITMQ_ENV = ""  # 生产环境
-RABBITMQ_ENV = os.getenv("RABBITMQ_ENV", "-dev")
-# RABBITMQ_ENV = "-local"  # 本地测试环境
-
-if RABBITMQ_ENV == "-dev":
-    JAVA_STREAM_API_URL = f"https://develop.api.aida.com.hk/api/third/party/receiveDesignResults"
-elif RABBITMQ_ENV == "-prod":
-    JAVA_STREAM_API_URL = f"https://api.aida.com.hk/api/third/party/receiveDesignResults"
-
-settings = Settings()
-
-# minio 配置
-MINIO_URL = "www.minio-api.aida.com.hk"
-MINIO_ACCESS = 'vXKFLSJkYeEq2DrSZvkB'
-MINIO_SECRET = 'uKTZT3x7C43WvPN9QTc99DiRkwddWZrG9Uh3JVlR'
-MINIO_SECURE = True
-
-# S3 配置
-S3_ACCESS_KEY = "AKIAVD3OJIMF6UJFLSHZ"
-S3_AWS_SECRET_ACCESS_KEY = "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"
-S3_REGION_NAME = "ap-east-1"
-
-# redis 配置
-REDIS_HOST = "10.1.1.240"
-REDIS_PORT = "6379"
-REDIS_DB = "2"
-
-# rabbitmq config
-RABBITMQ_PARAMS = {
-    "host": "18.167.251.121",
-    "port": 5672,
-    "credentials": pika.credentials.PlainCredentials(username='rabbit', password='123456'),
-    "virtual_host": "/"
-}
-
-# milvus 配置
-MILVUS_URL = "http://10.1.1.240:19530"
-MILVUS_TOKEN = "root:Milvus"
-MILVUS_ALIAS = "default"
-MILVUS_TABLE_KEYPOINT = "keypoint_cache_2"
-MILVUS_TABLE_SEG = "seg_cache"
-
-# Mysql 配置
-DB_HOST = '18.167.251.121'  # 数据库主机地址
-# DB_PORT = int( 33006)
-DB_PORT = 33008  # 数据库端口
-DB_USERNAME = 'aida_con_python'  # 数据库用户名
-DB_PASSWORD = '123456'  # 数据库密码
-DB_NAME = 'aida'  # 数据库库名
-
-# openai
-os.environ['SERPAPI_API_KEY'] = "a793513017b0718db7966207c31703d280d12435c982f1e67bbcbffa52e7632c"
-OPENAI_STREAM = True
-BUFFER_THRESHOLD = 6  # must be even number
-SINGLE_TOKEN_THRESHOLD = 200
-TOKEN_THRESHOLD = 600
-OPENAI_TEMPERATURE = 0
-
-# OPENAI_API_KEY =  "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"
-OPENAI_API_KEY = "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"
-OPENAI_MODEL = "gpt-3.5-turbo-0613"
-OPENAI_MODEL_LIST = {"gpt-3.5-turbo-0613",
-                     "gpt-3.5-turbo-16k-0613",
-                     "gpt-4-0314",
-                     "gpt-4-32k-0314",
-                     "gpt-4-0613",
-                     "gpt-4-32k-0613", }
-
-# SR service config
-SR_MODEL_NAME = "super_resolution"
-SR_TRITON_URL = "10.1.1.240:10031"
-SR_MINIO_BUCKET = "aida-users"
-SR_RABBITMQ_QUEUES = f"SuperResolution{RABBITMQ_ENV}"
-
-# GenerateImage service config
-FAST_GI_MODEL_URL = '10.1.1.243:10011'
-FAST_GI_MODEL_NAME = 'stable_diffusion_xl'
-
-GI_MODEL_URL = '10.1.1.240:10061'
-GI_MODEL_NAME = 'flux'
-
-GMV_MODEL_URL = '10.1.1.243:10081'
-GMV_MODEL_NAME = 'multi_view'
-
-GMV_RABBITMQ_QUEUES = f"GenerateMultiView{RABBITMQ_ENV}"
-
-GI_MINIO_BUCKET = "aida-users"
-GI_RABBITMQ_QUEUES = f"GenerateImage{RABBITMQ_ENV}"
-GI_SYS_IMAGE_URL = "aida-sys-image/generate_image/white_image.jpg"
-
-# SLOGAN service config
-SLOGAN_RABBITMQ_QUEUES = f"Slogan{RABBITMQ_ENV}"
-
-# Generate Single Logo service config
-GSL_MODEL_URL = '10.1.1.243:10041'
-GSL_MINIO_BUCKET = "aida-users"
-GSL_MODEL_NAME = 'stable_diffusion_xl_transparent'
-GEN_SINGLE_LOGO_RABBITMQ_QUEUES = f"GenSingleLogo{RABBITMQ_ENV}"
-
-# Generate Product service config
-# GPI_RABBITMQ_QUEUES = os.getenv("GEN_PRODUCT_IMAGE_RABBITMQ_QUEUES", f"ToProductImage{RABBITMQ_ENV}")
-# GPI_MODEL_NAME_OVERALL = 'sdxl_ensemble_all'
-# GPI_MODEL_URL = '10.1.1.243:10051'
-
-# Generate Product service config 旧版product img 模型
-GPI_RABBITMQ_QUEUES = f"ToProductImage{RABBITMQ_ENV}"
-BATCH_GPI_RABBITMQ_QUEUES = f"BatchToProductImage{RABBITMQ_ENV}"
-GPI_MODEL_NAME_OVERALL = 'diffusion_ensemble_all'
-GPI_MODEL_NAME_SINGLE = 'stable_diffusion_1_5_cnet'
-GPI_MODEL_URL = '10.1.1.243:10051'
-
-# Generate Single Logo service config
-GRI_RABBITMQ_QUEUES = f"Relight{RABBITMQ_ENV}"
-BATCH_GRI_RABBITMQ_QUEUES = f"BatchRelight{RABBITMQ_ENV}"
-GRI_MODEL_NAME_OVERALL = 'diffusion_relight_ensemble'
-GRI_MODEL_NAME_SINGLE = 'stable_diffusion_1_5_relight'
-GRI_MODEL_URL = '10.1.1.240:10051'
-
-# Pose Transform service config
-
-PS_RABBITMQ_QUEUES = f"PoseTransform{RABBITMQ_ENV}"
-BATCH_PS_RABBITMQ_QUEUES = f"BatchPoseTransform{RABBITMQ_ENV}"
-PT_MODEL_URL = '10.1.1.243:10061'
-
-# SEG service config
-SEGMENTATION = {
-    "new_model_name": "seg_knet",
-    "name": "seg_ocrnet_hr18",
-    "input": "seg_input__0",
-    "output": "seg_output__0",
-}
-# ollama config
-OLLAMA_URL = "http://10.1.1.240:11434/api/embeddings"
-
-# design batch
-BATCH_DESIGN_RABBITMQ_QUEUES = f"DesignBatch{RABBITMQ_ENV}"
-
-# DESIGN config
-DESIGN_MODEL_URL = '10.1.1.240:10000'
-AIDA_CLOTHING = "aida-clothing"
-KEYPOINT_RESULT_TABLE_FIELD_SET = ('neckline_left', 'neckline_right', 'shoulder_left', 'shoulder_right', 'armpit_left', 'armpit_right',
-                                   'cuff_left_in', 'cuff_left_out', 'cuff_right_in', 'cuff_right_out', 'waistband_left', 'waistband_right')
-
-# DESIGN 预处理
-IF_DEBUG_SHOW = False
-
-# 优先级
-PRIORITY_DICT = {
-    'earring_front': 99,
-    'bag_front': 98,
-    'hairstyle_front': 97,
-    'outwear_front': 20,
-    'tops_front': 19,
-    'dress_front': 18,
-    'blouse_front': 17,
-    'skirt_front': 16,
-    'trousers_front': 15,
-    'bottoms_front': 14,
-    'shoes_right': 1,
-    'shoes_left': 1,
-    'body': 0,
-    'bottoms_back': -14,
-    'trousers_back': -15,
-    'skirt_back': -16,
-    'blouse_back': -17,
-    'dress_back': -18,
-    'tops_back': -19,
-    'outwear_back': -20,
-    'hairstyle_back': -97,
-    'bag_back': -98,
-    'earring_back': -99,
-}
-
-QWEN_API_KEY = "sk-f31c29e61ac2498ba5e307aaa6dc10e0"
-
-DB_CONFIG = {
-    "host": "18.167.251.121",
-    "port": 3306,
-    "user": "root",
-    "password": "QWa998345",
-    "database": "aida",
-    "charset": "utf8mb4"
-}
-
-TABLE_CATEGORIES = {
-    "female_dress": "female/dress",
-    "female_outwear": "female/outwear",
-    "female_trousers": "female/trousers",
-    "female_skirt": "female/skirt",
-    "female_blouse": "female/blouse",
-    "male_tops": "male/tops",
-    "male_bottoms": "male/bottoms",
-    "male_outwear": "male/outwear"
-}
-
-# --- ComfyUI 配置信息 ---
-COMFYUI_SERVER_ADDRESS = "10.1.2.227:8080"  # 替换为您的 ComfyUI 服务器地址

app/core/config.py

@@ -36,7 +36,7 @@ class Settings(BaseSettings):
 
     # --- mysql 配置信息 ---
     MYSQL_HOST: str = Field(default='', description="")
-    MYSQL_PORT: int = Field(default='', description="")
+    MYSQL_PORT: int = Field(default=3306, description="")
     MYSQL_USER: str = Field(default='', description="")
     MYSQL_PASSWORD: str = Field(default='', description="")
     MYSQL_DB: str = Field(default='', description="")
@@ -64,11 +64,19 @@ class Settings(BaseSettings):
     # --- Design Callback Java 接口 ---
     JAVA_STREAM_API_URL: str = Field(default='', description="")
 
+    # ---  flux2 klein model url ---
+    FLUX2_GEN_IMG_MODEL_URL: str = Field(default='', description="")
+
+    # --- 服务器IP ---
+    A6000_SERVICE_HOST: str = Field(default='', description="")
+    B_4_X_4090_SERVICE_HOST: str = Field(default='', description="")
+
     # --- 其他配置信息 以下均为Docker容器内配置---
     LOGS_PATH: str = Field(default="/logs/", description="")
     CATEGORY_PATH: str = Field(default="/app/service/attribute/config/descriptor/category/category_dis.csv", description="")
     SEG_CACHE_PATH: str = Field(default="/seg_cache/", description="")
     RECOMMEND_PATH_PREFIX: str = Field(default="/app/service/recommend/", description="")
+    SERVE_PORT: int = Field(default=2010, description="")
 
 
 settings = Settings()
@@ -117,39 +125,41 @@ KEYPOINT_RESULT_TABLE_FIELD_SET = ('neckline_left', 'neckline_right', 'shoulder_
 MILVUS_TABLE_KEYPOINT = "keypoint_cache_2"
 
 # ollama 地址
-OLLAMA_URL = "http://10.1.1.240:11434/api/embeddings"
+OLLAMA_URL = f"http://{settings.A6000_SERVICE_HOST}:11434/api/embeddings"
 
 """Triton Server Config"""
 # Design
-DESIGN_MODEL_URL = '10.1.1.240:10000'
+DESIGN_MODEL_URL = f'{settings.A6000_SERVICE_HOST}:10000'
 DESIGN_MODEL_NAME = 'seg_knet'
+# Seg Product
+SEG_PRODUCT_MODEL_URL = f'{settings.B_4_X_4090_SERVICE_HOST}:30000'
 # Generate Image
-GI_MODEL_URL = '10.1.1.240:10061'
+GI_MODEL_URL = f'{settings.A6000_SERVICE_HOST}:10061'
 GI_MODEL_NAME = 'flux'
 # Generate Single Logo
-GSL_MODEL_URL = '10.1.1.243:10041'
+GSL_MODEL_URL = f'{settings.B_4_X_4090_SERVICE_HOST}:10041'
 GSL_MODEL_NAME = 'stable_diffusion_xl_transparent'
 # Generate Product （整套和单品）
-GPI_MODEL_URL = '10.1.1.243:10051'
+GPI_MODEL_URL = f'{settings.B_4_X_4090_SERVICE_HOST}:10051'
 GPI_MODEL_NAME_OVERALL = 'diffusion_ensemble_all'
 GPI_MODEL_NAME_SINGLE = 'stable_diffusion_1_5_cnet'
 
 # 以下停用中...*************
 # 多视角生成
-GMV_MODEL_URL = '10.1.1.243:10081'
+GMV_MODEL_URL = f'{settings.B_4_X_4090_SERVICE_HOST}:10081'
 GMV_MODEL_NAME = 'multi_view'
 # 超分
 SR_MODEL_NAME = "super_resolution"
-SR_TRITON_URL = "10.1.1.240:10031"
+SR_TRITON_URL = f"{settings.A6000_SERVICE_HOST}:10031"
 # 打光
-GRI_MODEL_URL = '10.1.1.240:10051'
+GRI_MODEL_URL = f'{settings.A6000_SERVICE_HOST}:10051'
 GRI_MODEL_NAME_OVERALL = 'diffusion_relight_ensemble'
 GRI_MODEL_NAME_SINGLE = 'stable_diffusion_1_5_relight'
 # agent 图片生成
-FAST_GI_MODEL_URL = '10.1.1.243:10011'
+FAST_GI_MODEL_URL = f'{settings.B_4_X_4090_SERVICE_HOST}:10011'
 FAST_GI_MODEL_NAME = 'stable_diffusion_xl'
 # 图转视频 triton版
-PT_MODEL_URL = '10.1.1.243:10061'
+PT_MODEL_URL = f'{settings.B_4_X_4090_SERVICE_HOST}:10061'
 
 # *************
 

app/main.py

@@ -16,7 +16,7 @@ from fastapi.responses import JSONResponse
 
 from app.api.api_route import router
 from app.core.config import settings
-from app.core.record_api_count import count_api_calls
+# from app.core.record_api_count import count_api_calls
 from app.schemas.response_template import ResponseModel
 from logging_env import LOGGER_CONFIG_DICT
 from dotenv import load_dotenv
@@ -48,7 +48,7 @@ def get_application() -> FastAPI:
         allow_methods=["*"],
         allow_headers=["*"],
     )
-    application.middleware("http")(count_api_calls)
+    # application.middleware("http")(count_api_calls)
     application.include_router(router=router)
     return application
 

app/schemas/design.py

@@ -4,12 +4,13 @@ from pydantic import BaseModel, Field
 
 
 class SAMRequestModel(BaseModel):
-    user_id: int = Field(..., description="用户id, 必填字段")
+    bucket: str = Field(..., description="minio bucket name ")
+    object_name: str = Field(..., description="minio object name ")
     image_path: str = Field(..., description="图片路径，必填字段")
     type: str = Field(..., description="推理类型，必填字段")
-    points: Optional[List[List[float]]] = None
+    points: Optional[List[List[float]]] | None = None
-    labels: Optional[List[int]] = None
+    labels: Optional[List[int]] | None = None
-    box: Optional[List[int]] = None
+    box: Optional[List[int]] | None = None
 
 
 class DesignModel(BaseModel):

app/schemas/generate_image.py

@@ -1,6 +1,6 @@
-from typing import List
+from typing import List, Optional
 
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
 
 
 class GenerateMultiViewModel(BaseModel):
@@ -8,6 +8,17 @@ class GenerateMultiViewModel(BaseModel):
     image_url: str
 
 
+class GenerateImageFlux2KleinModel(BaseModel):
+    bucket_name: str = Field(..., description="OSS桶名，不传则为None")
+    object_name: str = Field(..., description="OSS对象名（文件路径），不传则为None")
+    # input_image_paths: Optional[List[str]] = Field(default=[], description="输入图片路径列表")
+    width: Optional[int] = Field(default=1024, description="图片宽度，默认512像素")
+    height: Optional[int] = Field(default=1024, description="图片高度，默认512像素")
+    prompt: Optional[str] = Field(default="", description="文本提示词，用于模型推理等场景")
+    steps: Optional[int] = Field(default=4, description="推理步数，控制模型生成过程的迭代次数")
+    guidance: Optional[float] = Field(default=4.0, description="引导系数，调节提示词对生成结果的影响程度")
+
+
 class GenerateImageModel(BaseModel):
     tasks_id: str
     prompt: str
@@ -24,6 +35,13 @@ class GenerateSingleLogoImageModel(BaseModel):
     seed: str
 
 
+class GenerateSloganImageModel(BaseModel):
+    num_point: int
+    tasks_id: str
+    prompt: str
+    image_url: str
+
+
 class GenerateProductImageModel(BaseModel):
     tasks_id: str
     prompt: str
@@ -32,6 +50,13 @@ class GenerateProductImageModel(BaseModel):
     product_type: str
 
 
+class Flux2ToProductImgModel(BaseModel):
+    tasks_id: str
+    prompt: str
+    image_path: str
+    infer_step: int | None = None
+
+
 class GenerateRelightImageModel(BaseModel):
     tasks_id: str
     prompt: str

app/service/attribute/service_att_recognition.py

@@ -3,7 +3,6 @@
 from pprint import pprint
 
 import cv2
-import mmcv
 import numpy as np
 import pandas as pd
 import torch
@@ -12,6 +11,7 @@ from minio import Minio
 
 from app.core.config import settings, DESIGN_MODEL_URL
 from app.schemas.attribute_retrieve import AttributeRecognitionModel
+from app.service.utils.image_normalize import my_imnormalize
 from app.service.utils.new_oss_client import oss_get_image
 
 minio_client = Minio(settings.MINIO_URL, access_key=settings.MINIO_ACCESS, secret_key=settings.MINIO_SECRET, secure=settings.MINIO_SECURE)
@@ -109,10 +109,9 @@ class AttributeRecognition:
 
     @staticmethod
     def preprocess(img):
-        img = mmcv.imread(img)
         img_scale = (224, 224)
         img = cv2.resize(img, img_scale)
-        img = mmcv.imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
+        img = my_imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
         preprocessed_img = np.expand_dims(img.transpose(2, 0, 1), axis=0)
         return preprocessed_img
 

app/service/attribute/service_category_recognition.py

@@ -10,7 +10,6 @@
 from minio import Minio
 from skimage import transform
 import cv2
-import mmcv
 import numpy as np
 import pandas as pd
 import tritonclient.http as httpclient
@@ -18,6 +17,7 @@ import torch
 
 from app.core.config import settings, DESIGN_MODEL_URL
 from app.schemas.attribute_retrieve import CategoryRecognitionModel
+from app.service.utils.image_normalize import my_imnormalize
 from app.service.utils.new_oss_client import oss_get_image
 
 minio_client = Minio(settings.MINIO_URL, access_key=settings.MINIO_ACCESS, secret_key=settings.MINIO_SECRET, secure=settings.MINIO_SECURE)
@@ -39,11 +39,10 @@ class CategoryRecognition:
 
     @staticmethod
     def preprocess(img):
-        img = mmcv.imread(img)
         # ori_shape = img.shape[:2]
         img_scale = (224, 224)
         img = cv2.resize(img, img_scale)
-        img = mmcv.imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
+        img = my_imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
         preprocessed_img = np.expand_dims(img.transpose(2, 0, 1), axis=0)
         return preprocessed_img
 

app/service/brand_dna/service.py

@@ -1,7 +1,6 @@
 import logging
 
 import cv2
-import mmcv
 import numpy as np
 import pandas as pd
 import torch
@@ -9,11 +8,12 @@ import torch.nn.functional as F
 import tritonclient.http as httpclient
 from minio import Minio
 
-from app.core.config import DESIGN_MODEL_URL
+from app.core.config import DESIGN_MODEL_URL, SEG_PRODUCT_MODEL_URL
 from app.core.config import settings
 from app.schemas.brand_dna import BrandDnaModel
 from app.service.attribute.config import const
 from app.service.utils.generate_uuid import generate_uuid
+from app.service.utils.image_normalize import my_imnormalize
 from app.service.utils.new_oss_client import oss_upload_image, oss_get_image
 
 minio_client = Minio(settings.MINIO_URL, access_key=settings.MINIO_ACCESS, secret_key=settings.MINIO_SECRET, secure=settings.MINIO_SECURE)
@@ -29,7 +29,7 @@ class BrandDna:
         self.attr_type = pd.read_csv(settings.CATEGORY_PATH)
         # self.attr_type = pd.read_csv(r"E:\workspace\trinity_client_aida\app\service\attribute\config\descriptor\category\category_dis.csv")
         self.att_client = httpclient.InferenceServerClient(url=DESIGN_MODEL_URL)
-        self.seg_client = httpclient.InferenceServerClient(url='10.1.1.243:30000')
+        self.seg_client = httpclient.InferenceServerClient(url=SEG_PRODUCT_MODEL_URL)
         self.const = const
         # self.const = local_debug_const
 
@@ -202,7 +202,7 @@ class BrandDna:
     # 服装分割预处理
     @staticmethod
     def seg_product_preprocess(image):
-        img = mmcv.imread(image)
+        img = image
         ori_shape = img.shape[:2]
         img_scale_w, img_scale_h = ori_shape
         if ori_shape[0] > 1024:
@@ -211,9 +211,9 @@ class BrandDna:
             img_scale_h = 1024
         # 如果图片size任意一边 大于 1024， 则会resize 成1024
         if ori_shape != (img_scale_w, img_scale_h):
-            # mmcv.imresize(img, img_scale_h, img_scale_w) # 老代码 引以为戒！哈哈哈~ h和w写反了
+            # my_imnormalize(img, img_scale_h, img_scale_w) # 老代码 引以为戒！哈哈哈~ h和w写反了
             img = cv2.resize(img, (img_scale_h, img_scale_w))
-        img = mmcv.imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
+        img = my_imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
         preprocessed_img = np.expand_dims(img.transpose(2, 0, 1), axis=0)
         return preprocessed_img, ori_shape
 
@@ -227,11 +227,10 @@ class BrandDna:
     # 类别检测模型预处理
     @staticmethod
     def category_preprocess(img):
-        img = mmcv.imread(img)
         # ori_shape = img.shape[:2]
         img_scale = (224, 224)
         img = cv2.resize(img, img_scale)
-        img = mmcv.imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
+        img = my_imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
         preprocessed_img = np.expand_dims(img.transpose(2, 0, 1), axis=0)
         return preprocessed_img
 

app/service/brand_dna/service_generate_brand_info.py

@@ -1,19 +1,10 @@
-import logging
+import uuid
-
+import httpx
-import cv2
-import numpy as np
-import tritonclient.grpc as grpcclient
 from langchain_classic.output_parsers import ResponseSchema, StructuredOutputParser
 from langchain_community.chat_models import ChatTongyi
 from langchain_core.prompts import PromptTemplate
 from minio import Minio
-from tritonclient.utils import np_to_triton_dtype
-
-from app.core.config import GI_MODEL_URL, GI_MODEL_NAME
 from app.schemas.brand_dna import GenerateBrandModel
-from app.service.utils.generate_uuid import generate_uuid
-from app.service.utils.new_oss_client import oss_upload_image
-
 from app.core.config import settings
 
 
@@ -26,14 +17,9 @@ class GenerateBrandInfo:
         # user info init
         self.user_id = request_data.user_id
         self.category = "brand_logo"
-        # generate logo init
-        self.grpc_client = grpcclient.InferenceServerClient(url=GI_MODEL_URL)
-        self.image = np.random.randint(0, 256, (1024, 1024, 3), dtype=np.uint8)
-        self.batch_size = 1
-        self.mode = 'txt2img'
 
         # llm generate brand info init
-        self.model = ChatTongyi(model="qwen2.5-14b-instruct", api_key="sk-7658298c6b99443c98184a5e634fe6ab")
+        self.model = ChatTongyi(model="qwen2.5-14b-instruct", api_key=settings.QWEN_API_KEY)
 
         self.response_schemas = [
             ResponseSchema(name="brand_name", description="Brand name."),
@@ -63,38 +49,20 @@ class GenerateBrandInfo:
         self.generate_logo_prompt = brand_data['brand_logo_prompt']
 
     def generate_brand_logo(self):
-        prompts = [self.generate_logo_prompt] * self.batch_size
+        request_item = {
-        modes = [self.mode] * self.batch_size
+            "bucket_name": "aida-users",
-        images = [self.image.astype(np.float16)] * self.batch_size
+            "object_name": f'{self.user_id}/{self.category}/{uuid.uuid4().hex}.png',
-
+            "prompt": self.generate_logo_prompt,
-        text_obj = np.array(prompts, dtype="object").reshape((-1, 1))
+            "height": 1024,
-        mode_obj = np.array(modes, dtype="object").reshape((-1, 1))
+            "width": 1024
-        image_obj = np.array(images, dtype=np.float16).reshape((-1, 1024, 1024, 3))
+        }
-
+        with httpx.Client(timeout=120) as client:
-        input_text = grpcclient.InferInput("prompt", text_obj.shape, np_to_triton_dtype(text_obj.dtype))
+            resp = client.post(
-        input_image = grpcclient.InferInput("input_image", image_obj.shape, np_to_triton_dtype(image_obj.dtype))
+                f"http://{settings.FLUX2_GEN_IMG_MODEL_URL}/predict",
-        input_mode = grpcclient.InferInput("mode", mode_obj.shape, np_to_triton_dtype(mode_obj.dtype))
+                json=request_item,
-
+            )
-        input_text.set_data_from_numpy(text_obj)
+        result = resp.json()
-        input_image.set_data_from_numpy(image_obj)
+        self.result_data['brand_logo'] = result.get("output_path", "")
-        input_mode.set_data_from_numpy(mode_obj)
-
-        inputs = [input_text, input_image, input_mode]
-        result = self.grpc_client.infer(model_name=GI_MODEL_NAME, inputs=inputs)
-        image = result.as_numpy("generated_image")
-        image_result = cv2.cvtColor(np.squeeze(image.astype(np.uint8)), cv2.COLOR_RGB2BGR)
-        logo_url = self.upload_logo_image(image_result, generate_uuid())
-        self.result_data['brand_logo'] = logo_url
-
-    def upload_logo_image(self, image, object_name):
-        try:
-            _, img_byte_array = cv2.imencode('.jpg', image)
-            object_name = f'{self.user_id}/{self.category}/{object_name}.jpg'
-            oss_upload_image(oss_client=self.minio_client, bucket="aida-users", object_name=object_name, image_bytes=img_byte_array)
-            image_url = f"aida-users/{object_name}"
-            return image_url
-        except Exception as e:
-            logging.warning(f"upload_png_mask runtime exception : {e}")
 
 
 if __name__ == '__main__':

app/service/clothing_seg/service.py

@@ -23,7 +23,7 @@ class ClothingSeg:
     def __init__(self, request_data):
         self.image_data = request_data.image_data
         self.user_id = request_data.user_id
-        self.triton_client = grpcclient.InferenceServerClient(url="10.1.1.243:10071")
+        self.triton_client = grpcclient.InferenceServerClient(url=f"{settings.B_4_X_4090_SERVICE_HOST}:10071")
 
     @RunTime
     def get_result(self):
@@ -139,7 +139,7 @@ def get_bounding_box(mask):
 
 if __name__ == "__main__":
     test_data = ClothingSegModel(
-        user_id=89,
+        user_id="89",
         image_data=[
             # {
             #     "image_url": "test/clothing_seg/dress.jpg",

app/service/comfyui_I2V/pose2v_server.py

@@ -13,7 +13,7 @@ from PIL import Image
 from minio import Minio, S3Error
 from moviepy.video.io.VideoFileClip import VideoFileClip
 
-from app.core.config import  settings
+from app.core.config import settings, PS_RABBITMQ_QUEUES
 from app.schemas.comfyui_i2v import ComfyuiPose2VModel
 from app.service.generate_image.utils.mq import publish_status
 
@@ -622,9 +622,9 @@ class ComfyUIServerPose2V:
 
             # 推送消息
             if not settings.DEBUG:
-                publish_status(json.dumps(self.pose_transform_data), settings.COMFYUI_SERVER_ADDRESS)
+                publish_status(json.dumps(self.pose_transform_data), PS_RABBITMQ_QUEUES)
             logger.info(
-                f" [x] Sent to： {settings.COMFYUI_SERVER_ADDRESS} data：@@@@  {json.dumps(self.pose_transform_data, indent=4)}")
+                f" [x] Sent to： {PS_RABBITMQ_QUEUES} data：@@@@  {json.dumps(self.pose_transform_data, indent=4)}")
 
             return "\n🎉 所有任务完成！"
 

app/service/design_batch/utils/design_ensemble.py

@@ -10,13 +10,13 @@
 import logging
 
 import cv2
-import mmcv
 import numpy as np
 import torch
 import torch.nn.functional as F
 import tritonclient.http as httpclient
 
 from app.core.config import DESIGN_MODEL_URL, DESIGN_MODEL_NAME
+from app.service.utils.image_normalize import my_imnormalize
 
 """
     keypoint
@@ -25,13 +25,13 @@ from app.core.config import DESIGN_MODEL_URL, DESIGN_MODEL_NAME
 
 
 def keypoint_preprocess(img_path):
-    img = mmcv.imread(img_path)
+    img = img_path
     img_scale = (256, 256)
     h, w = img.shape[:2]
     img = cv2.resize(img, img_scale)
     w_scale = img_scale[0] / w
     h_scale = img_scale[1] / h
-    img = mmcv.imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
+    img = my_imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
     preprocessed_img = np.expand_dims(img.transpose(2, 0, 1), axis=0)
     return preprocessed_img, (w_scale, h_scale)
 
@@ -74,7 +74,7 @@ def keypoint_postprocess(output, scale_factor):
 
 # KNet
 def seg_preprocess(img_path):
-    img = mmcv.imread(img_path)
+    img = img_path
     ori_shape = img.shape[:2]
     img_scale_w, img_scale_h = ori_shape
     if ori_shape[0] > 1024:
@@ -83,9 +83,9 @@ def seg_preprocess(img_path):
         img_scale_h = 1024
     # 如果图片size任意一边 大于 1024， 则会resize 成1024
     if ori_shape != (img_scale_w, img_scale_h):
-        # mmcv.imresize(img, img_scale_h, img_scale_w) # 老代码 引以为戒！哈哈哈~ h和w写反了
+        # my_imnormalize(img, img_scale_h, img_scale_w) # 老代码 引以为戒！哈哈哈~ h和w写反了
         img = cv2.resize(img, (img_scale_h, img_scale_w))
-    img = mmcv.imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
+    img = my_imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
     preprocessed_img = np.expand_dims(img.transpose(2, 0, 1), axis=0)
     return preprocessed_img, ori_shape
 

app/service/design_fast/design_generate.py

@@ -130,7 +130,7 @@ def design_generate(request_data):
                 items_response['synthesis_url'] = synthesis(layers, new_size, basic)
 
         else:
-            item_result = process_item(object['items'][0], basic)
+            item_result = process_item(object['items'][0], basic, design_type)
             items_response['layers'].append({
                 'image_category': f"{item_result['name']}_front",
                 'image_size': item_result['back_image'].size if item_result['back_image'] else None,
@@ -184,6 +184,7 @@ def design_generate_v2(request_data):
 
     def process_object(object, callback_url):
         basic = object['basic']
+        design_type = basic.get('design_type', "default")
         items_response = {
             'layers': [],
             'objectSign': object['objectSign'] if 'objectSign' in object.keys() else "",
@@ -192,7 +193,7 @@ def design_generate_v2(request_data):
         if basic['single_overall'] == "overall":
             item_results = []
             for item in object['items']:
-                item_results.append(process_item(item, basic))
+                item_results.append(process_item(item, basic, design_type))
             layers = []
             for item in item_results:
                 process_layer(item, layers)
@@ -217,7 +218,7 @@ def design_generate_v2(request_data):
                 })
             items_response['synthesis_url'] = synthesis(layers, new_size, basic)
         else:
-            item_result = process_item(object['items'][0], basic)
+            item_result = process_item(object['items'][0], basic, design_type)
             items_response['layers'].append({
                 'image_category': f"{item_result['name']}_front",
                 'image_size': item_result['back_image'].size if item_result['back_image'] else None,

app/service/design_fast/item.py

@@ -16,6 +16,9 @@ class OthersItem(BaseItem):
         self.Others_pipeline = [
             LoadImage(minio_client),
             Segmentation(minio_client),
+            Color(minio_client),
+            NoSegPrintPainting(minio_client),
+            PrintPainting(minio_client),
             Scaling(),
             Split(minio_client)
         ]
@@ -82,8 +85,8 @@ class OthersMergeItem(BaseItem):
             Segmentation(minio_client),
             # BackPerspective(minio_client),
             Color(minio_client),
-            NoSegPrintPainting(minio_client),
+            # NoSegPrintPainting(minio_client),
-            PrintPainting(minio_client),
+            # PrintPainting(minio_client),
             Scaling(),
             Split(minio_client)
         ]

app/service/design_fast/pipeline/no_seg_print_painting.py

@@ -12,9 +12,13 @@ class NoSegPrintPainting:
         self.minio_client = minio_client
 
     def __call__(self, result):
-        single_print = result['print']['single']
+        # single_print = [result['print']['single']]
         overall_print = result['print']['overall']
-        element_print = result['print']['element']
+        # element_print = result['print']['element'
+
+        single_print = None
+        element_print = None
+
         result['single_image'] = None
         result['print_image'] = None
 
@@ -23,9 +27,9 @@ class NoSegPrintPainting:
             # 获取平铺 + 旋转 的overall print
             painting_dict = self.painting_collection(painting_dict, overall_print)
             result['no_seg_sketch_overall'] = result['no_seg_sketch_print'] = self.printpaint(result, painting_dict, print_=True)
-            result['pattern_image'] = result['no_seg_sketch_overall']
+            # result['pattern_image'] = result['no_seg_sketch_overall']
 
-        if single_print['print_path_list']:
+        if single_print:
             print_background = np.zeros((result['pattern_image'].shape[0], result['pattern_image'].shape[1], 3), dtype=np.uint8)
             mask_background = np.zeros((result['pattern_image'].shape[0], result['pattern_image'].shape[1], 3), dtype=np.uint8)
             for i in range(len(single_print['print_path_list'])):
@@ -65,7 +69,7 @@ class NoSegPrintPainting:
                 single_image = cv2.add(tmp1, tmp2)
                 result['no_seg_sketch_print'] = single_image
 
-        if element_print['element_path_list']:
+        if element_print:
             print_background = np.zeros((result['final_image'].shape[0], result['final_image'].shape[1], 3), dtype=np.uint8)
             mask_background = np.zeros((result['final_image'].shape[0], result['final_image'].shape[1], 3), dtype=np.uint8)
             for i in range(len(element_print['element_path_list'])):
@@ -162,15 +166,17 @@ class NoSegPrintPainting:
         dim_max = max(painting_dict['dim_image_h'], painting_dict['dim_image_w'])
         dim_pattern = (int(dim_max * print_['scale'] / 5), int(dim_max * print_['scale'] / 5))
         gap = print_dict.get('gap', [[0, 0]])[0]
-        painting_dict['tile_print'] = tile_image(pattern=print_['image'],
+        painting_dict['tile_print'], painting_dict['mask_inv_print'] = tile_image(pattern=print_['image'],
-                                                 dim=dim_pattern,
+                                                                                  mask=print_['mask'],
-                                                 gap_x=gap[0],
+                                                                                  dim=dim_pattern,
-                                                 gap_y=gap[1],
+                                                                                  gap_x=gap[0],
-                                                 canvas_h=painting_dict['dim_image_h'],
+                                                                                  gap_y=gap[1],
-                                                 canvas_w=painting_dict['dim_image_w'],
+                                                                                  canvas_h=painting_dict['dim_image_h'],
-                                                 location=painting_dict['location'],
+                                                                                  canvas_w=painting_dict['dim_image_w'],
-                                                 angle=45)
+                                                                                  location=painting_dict['location'],
-        painting_dict['mask_inv_print'] = np.zeros(painting_dict['tile_print'].shape[:2], dtype=np.uint8)
+                                                                                  angle=int(print_.get('print_angle_list', [0])[0]))
+        # painting_dict['mask_inv_print'] = np.zeros(painting_dict['tile_print'].shape[:2], dtype=np.uint8)
+        # painting_dict['mask_inv_print'] = self.get_mask_inv(painting_dict['tile_print'])
         return painting_dict
 
     def tile_image(self, pattern, dim, scale, dim_image_h, dim_image_w, location, trigger=False):
@@ -251,10 +257,15 @@ class NoSegPrintPainting:
         image = oss_get_image(oss_client=self.minio_client, bucket=bucket_name, object_name=object_name, data_type="PIL")
         # 判断图片格式，如果是RGBA 则贴在一张纯白图片上 防止透明转黑
         if image.mode == "RGBA":
+            mask_pil = image.split()[3]
             new_background = Image.new('RGB', image.size, (255, 255, 255))
             new_background.paste(image, mask=image.split()[3])
             image = new_background
+        else:
+            mask_pil = Image.new('L', image.size, 255)  # L=灰度图，255=纯白
         print_dict['image'] = cv2.cvtColor(np.asarray(image), cv2.COLOR_RGB2BGR)
+        print_dict['mask'] = cv2.threshold(np.array(mask_pil), 127, 255, cv2.THRESH_BINARY)[1]
+
         return print_dict
 
     def crop_image(self, image, image_size_h, image_size_w, location, print_shape):
@@ -404,9 +415,12 @@ class NoSegPrintPainting:
         return cropped_img
 
 
-def tile_image(pattern, dim, gap_x, gap_y, canvas_h, canvas_w, location, angle=0):
+def tile_image(pattern, mask, dim, gap_x, gap_y, canvas_h, canvas_w, location, angle=0):
     """
     按照指定的 X/Y 间距平铺印花，并支持旋转
+    【修改版】以被平铺图案的【中心】作为平铺基准点
+
+    :param location: [[center_y, center_x]]  → 第一个图案中心的坐标
     :param angle: 旋转角度 (度数, 逆时针)
     """
     # 1. 确保输入是 RGBA
@@ -418,35 +432,54 @@ def tile_image(pattern, dim, gap_x, gap_y, canvas_h, canvas_w, location, angle=0
     rotated_p = rotate_image(resized_p, angle)
     p_h, p_w = rotated_p.shape[:2]
 
-    # 3. 创建透明单元格
+    # 3. 创建透明单元格（图案放在单元格中心）
-    cell_h, cell_w = p_h + gap_y, p_w + gap_x
+    cell_h = p_h + gap_y
+    cell_w = p_w + gap_x
+
     unit_cell = np.zeros((cell_h, cell_w, 4), dtype=np.uint8)
-    unit_cell[:p_h, :p_w, :] = rotated_p
+
+    # 计算图案在单元格中的左上角位置（让图案居中）
+    start_y = (cell_h - p_h) // 2
+    start_x = (cell_w - p_w) // 2
+    unit_cell[start_y:start_y + p_h, start_x:start_x + p_w, :] = rotated_p
 
     # 4. 执行平铺
-    tiles_y = (canvas_h // cell_h) + 2
+    tiles_y = (canvas_h // cell_h) + 3  # 多加一点余量更安全
-    tiles_x = (canvas_w // cell_w) + 2
+    tiles_x = (canvas_w // cell_w) + 3
     full_tiled = np.tile(unit_cell, (tiles_y, tiles_x, 1))
 
-    # 5. 裁剪平铺层
+    # 5. 计算偏移（关键修改：以中心为基准）
-    offset_x = int(location[0][1] % cell_w)
+    center_y, center_x = location[0][0], location[0][1]  # 第一个图案的中心位置
-    offset_y = int(location[0][0] % cell_h)
+
+    # 计算从哪个位置开始裁剪，才能让中心落在指定坐标
+    offset_y = int((center_y - (p_h // 2)) % cell_h)
+    offset_x = int((center_x - (p_w // 2)) % cell_w)
+
     tiled_layer = full_tiled[offset_y: offset_y + canvas_h,
     offset_x: offset_x + canvas_w]
 
-    # 6. 创建纯白色背景并合成
+    # 6. 创建纯白色背景并合成（保持你原来的风格）
-    # 创建一个纯白色的 BGR 画布
     white_background = np.full((canvas_h, canvas_w, 3), 255, dtype=np.uint8)
 
-    # 分离平铺层的颜色通道和 Alpha 通道
     tiled_bgr = tiled_layer[:, :, :3]
-    alpha_mask = tiled_layer[:, :, 3] / 255.0  # 归一化到 0-1
+    alpha_mask = tiled_layer[:, :, 3] / 255.0
-    alpha_mask = cv2.merge([alpha_mask, alpha_mask, alpha_mask])  # 扩展到 3 通道
+    alpha_mask = cv2.merge([alpha_mask, alpha_mask, alpha_mask])
 
-    # 执行 Alpha 混合：结果 = 平铺层 * alpha + 背景 * (1 - alpha)
+    tiled_print = (tiled_bgr * alpha_mask + white_background * (1 - alpha_mask)).astype(np.uint8)
-    result = (tiled_bgr * alpha_mask + white_background * (1 - alpha_mask)).astype(np.uint8)
 
-    return result
+    # ====================== 处理 Mask ======================
+    # Mask 也同样居中处理
+    resized_mask = cv2.resize(mask, dim, interpolation=cv2.INTER_NEAREST)
+    rotated_mask = rotate_image(resized_mask, angle)  # 注意：mask也需要旋转
+
+    unit_mask = np.zeros((cell_h, cell_w), dtype=np.uint8)
+    unit_mask[start_y:start_y + p_h, start_x:start_x + p_w] = rotated_mask
+
+    full_mask_tiled = np.tile(unit_mask, (tiles_y, tiles_x))
+    tiled_mask = full_mask_tiled[offset_y: offset_y + canvas_h,
+    offset_x: offset_x + canvas_w]
+
+    return tiled_print, cv2.bitwise_not(tiled_mask)
 
 
 def rotate_image(image, angle):

app/service/design_fast/pipeline/print_painting.py

@@ -12,10 +12,14 @@ class PrintPainting:
         self.minio_client = minio_client
 
     def __call__(self, result):
-        single_print = result['print']['single']
+        # single_print = result['print']['single']
         overall_print = result['print']['overall']
-        element_print = result['print']['element']
+        # element_print = result['print']['element']
-        partial_path = result['print']['partial'] if 'partial' in result['print'] else None
+        # partial_path = result['print']['partial'] if 'partial' in result['print'] else None
+
+        single_print = None
+        element_print = None
+        partial_path = None
         result['single_image'] = None
         result['print_image'] = None
         # TODO 给result['pattern_image'] resize 到resize_scale的大小
@@ -37,13 +41,13 @@ class PrintPainting:
         if overall_print['print_path_list']:
             overall_print['location'][0] = [x * y for x, y in zip(overall_print['location'][0], result['resize_scale'])]
             painting_dict = {'dim_image_h': result['pattern_image'].shape[0], 'dim_image_w': result['pattern_image'].shape[1]}
-            result['print_image'] = result['pattern_image']
+            result['print_image'] = result['pattern_image'].copy()
             # 获取平铺 + 旋转 的overall print
             painting_dict = self.painting_collection(painting_dict, overall_print)
             result['print_image'] = self.printpaint(result, painting_dict, print_=True)
             result['single_image'] = result['final_image'] = result['pattern_image'] = result['print_image']
 
-        if single_print['print_path_list']:
+        if single_print:
             # 2025-9-19 印花调整 印花坐标按照sketch的缩放比调整
             sketch_resize_scale = result['resize_scale']
             print_background = np.zeros((result['pattern_image'].shape[0], result['pattern_image'].shape[1], 3), dtype=np.uint8)
@@ -84,7 +88,7 @@ class PrintPainting:
                 tmp2 = (result['final_image'] * (temp_fg / 255)).astype(np.uint8)
                 result['single_image'] = cv2.add(tmp1, tmp2)
 
-        if element_print['element_path_list']:
+        if element_print:
             # 2025-9-19 印花调整 印花坐标按照sketch的缩放比调整
             sketch_resize_scale = result['resize_scale']
             print_background = np.zeros((result['final_image'].shape[0], result['final_image'].shape[1], 3), dtype=np.uint8)
@@ -225,15 +229,15 @@ class PrintPainting:
         dim_max = max(painting_dict['dim_image_h'], painting_dict['dim_image_w'])
         dim_pattern = (int(dim_max * print_['scale'] / 5), int(dim_max * print_['scale'] / 5))
         gap = print_dict.get('gap', [[0, 0]])[0]
-        painting_dict['tile_print'] = tile_image(pattern=print_['image'],
+        painting_dict['tile_print'], painting_dict['mask_inv_print'] = tile_image(pattern=print_['image'],
-                                                 dim=dim_pattern,
+                                                                                  mask=print_['mask'],
-                                                 gap_x=gap[0],
+                                                                                  dim=dim_pattern,
-                                                 gap_y=gap[1],
+                                                                                  gap_x=gap[0],
-                                                 canvas_h=painting_dict['dim_image_h'],
+                                                                                  gap_y=gap[1],
-                                                 canvas_w=painting_dict['dim_image_w'],
+                                                                                  canvas_h=painting_dict['dim_image_h'],
-                                                 location=painting_dict['location'],
+                                                                                  canvas_w=painting_dict['dim_image_w'],
-                                                 angle=45)
+                                                                                  location=painting_dict['location'],
-        painting_dict['mask_inv_print'] = np.zeros(painting_dict['tile_print'].shape[:2], dtype=np.uint8)
+                                                                                  angle=int(print_.get('print_angle_list', [0])[0]))
         return painting_dict
 
     def tile_image(self, pattern, dim, scale, dim_image_h, dim_image_w, location, trigger=False):
@@ -314,10 +318,15 @@ class PrintPainting:
         image = oss_get_image(oss_client=self.minio_client, bucket=bucket_name, object_name=object_name, data_type="PIL")
         # 判断图片格式，如果是RGBA 则贴在一张纯白图片上 防止透明转黑
         if image.mode == "RGBA":
+            mask_pil = image.split()[3]
             new_background = Image.new('RGB', image.size, (255, 255, 255))
             new_background.paste(image, mask=image.split()[3])
             image = new_background
+        else:
+            mask_pil = Image.new('L', image.size, 255)  # L=灰度图，255=纯白
         print_dict['image'] = cv2.cvtColor(np.asarray(image), cv2.COLOR_RGB2BGR)
+        print_dict['mask'] = cv2.threshold(np.array(mask_pil), 127, 255, cv2.THRESH_BINARY)[1]
+
         return print_dict
 
     def crop_image(self, image, image_size_h, image_size_w, location, print_shape):
@@ -467,9 +476,12 @@ class PrintPainting:
         return cropped_img
 
 
-def tile_image(pattern, dim, gap_x, gap_y, canvas_h, canvas_w, location, angle=0):
+def tile_image(pattern, mask, dim, gap_x, gap_y, canvas_h, canvas_w, location, angle=0):
     """
     按照指定的 X/Y 间距平铺印花，并支持旋转
+    【修改版】以被平铺图案的【中心】作为平铺基准点
+
+    :param location: [[center_y, center_x]]  → 第一个图案中心的坐标
     :param angle: 旋转角度 (度数, 逆时针)
     """
     # 1. 确保输入是 RGBA
@@ -481,35 +493,54 @@ def tile_image(pattern, dim, gap_x, gap_y, canvas_h, canvas_w, location, angle=0
     rotated_p = rotate_image(resized_p, angle)
     p_h, p_w = rotated_p.shape[:2]
 
-    # 3. 创建透明单元格
+    # 3. 创建透明单元格（图案放在单元格中心）
-    cell_h, cell_w = p_h + gap_y, p_w + gap_x
+    cell_h = p_h + gap_y
+    cell_w = p_w + gap_x
+
     unit_cell = np.zeros((cell_h, cell_w, 4), dtype=np.uint8)
-    unit_cell[:p_h, :p_w, :] = rotated_p
+
+    # 计算图案在单元格中的左上角位置（让图案居中）
+    start_y = (cell_h - p_h) // 2
+    start_x = (cell_w - p_w) // 2
+    unit_cell[start_y:start_y + p_h, start_x:start_x + p_w, :] = rotated_p
 
     # 4. 执行平铺
-    tiles_y = (canvas_h // cell_h) + 2
+    tiles_y = (canvas_h // cell_h) + 3  # 多加一点余量更安全
-    tiles_x = (canvas_w // cell_w) + 2
+    tiles_x = (canvas_w // cell_w) + 3
     full_tiled = np.tile(unit_cell, (tiles_y, tiles_x, 1))
 
-    # 5. 裁剪平铺层
+    # 5. 计算偏移（关键修改：以中心为基准）
-    offset_x = int(location[0][1] % cell_w)
+    center_y, center_x = location[0][0], location[0][1]  # 第一个图案的中心位置
-    offset_y = int(location[0][0] % cell_h)
+
+    # 计算从哪个位置开始裁剪，才能让中心落在指定坐标
+    offset_y = int((center_y - (p_h // 2)) % cell_h)
+    offset_x = int((center_x - (p_w // 2)) % cell_w)
+
     tiled_layer = full_tiled[offset_y: offset_y + canvas_h,
     offset_x: offset_x + canvas_w]
 
-    # 6. 创建纯白色背景并合成
+    # 6. 创建纯白色背景并合成（保持你原来的风格）
-    # 创建一个纯白色的 BGR 画布
     white_background = np.full((canvas_h, canvas_w, 3), 255, dtype=np.uint8)
 
-    # 分离平铺层的颜色通道和 Alpha 通道
     tiled_bgr = tiled_layer[:, :, :3]
-    alpha_mask = tiled_layer[:, :, 3] / 255.0  # 归一化到 0-1
+    alpha_mask = tiled_layer[:, :, 3] / 255.0
-    alpha_mask = cv2.merge([alpha_mask, alpha_mask, alpha_mask])  # 扩展到 3 通道
+    alpha_mask = cv2.merge([alpha_mask, alpha_mask, alpha_mask])
 
-    # 执行 Alpha 混合：结果 = 平铺层 * alpha + 背景 * (1 - alpha)
+    tiled_print = (tiled_bgr * alpha_mask + white_background * (1 - alpha_mask)).astype(np.uint8)
-    result = (tiled_bgr * alpha_mask + white_background * (1 - alpha_mask)).astype(np.uint8)
 
-    return result
+    # ====================== 处理 Mask ======================
+    # Mask 也同样居中处理
+    resized_mask = cv2.resize(mask, dim, interpolation=cv2.INTER_NEAREST)
+    rotated_mask = rotate_image(resized_mask, angle)  # 注意：mask也需要旋转
+
+    unit_mask = np.zeros((cell_h, cell_w), dtype=np.uint8)
+    unit_mask[start_y:start_y + p_h, start_x:start_x + p_w] = rotated_mask
+
+    full_mask_tiled = np.tile(unit_mask, (tiles_y, tiles_x))
+    tiled_mask = full_mask_tiled[offset_y: offset_y + canvas_h,
+    offset_x: offset_x + canvas_w]
+
+    return tiled_print, cv2.bitwise_not(tiled_mask)
 
 
 def rotate_image(image, angle):

app/service/design_fast/pipeline/segmentation.py

@@ -47,9 +47,12 @@ class Segmentation:
                 # 本地查询seg 缓存是否存在
                 _, seg_result = self.load_seg_result(result["image_id"])
                 # 判断缓存和实际图片size是否相同
+                _ = False
                 if not _ or result["image"].shape[:2] != seg_result.shape:
                     # 推理获得seg 结果
                     seg_result = get_seg_result(result['image'])
+                    if result['name'] == 'others':
+                        seg_result = seg_result.clip(max=1)
                     self.save_seg_result(seg_result, result['image_id'])
             result['seg_result'] = seg_result
 

app/service/design_fast/pipeline/split.py

@@ -21,7 +21,9 @@ class Split(object):
         try:
             if result['name'] in ('outwear', 'dress', 'blouse', 'skirt', 'trousers', 'tops', 'bottoms', 'others'):
                 if result.get('design_type', None) == 'merge':
-                    # merge 不需要返回mask (红绿图)
+                    ori_front_mask = result['front_mask'].copy()
+                    ori_back_mask = result['back_mask'].copy()
+
                     if result['resize_scale'][0] == 1.0 and result['resize_scale'][1] == 1.0:
                         front_mask = result['front_mask']
                         back_mask = result['back_mask']
@@ -43,6 +45,20 @@ class Split(object):
                     result_front_image_pil = Image.fromarray(cv2.cvtColor(result_front_image, cv2.COLOR_BGR2RGBA))
                     result['front_image'], result["front_image_url"], _ = upload_png_mask(self.minio_client, result_front_image_pil, f'{generate_uuid()}', mask=None)
 
+                    height, width = ori_front_mask.shape
+                    mask_image = np.zeros((height, width, 3))
+                    mask_image[ori_front_mask != 0] = [0, 0, 255]
+                    mask_image[ori_back_mask != 0] = [0, 255, 0]
+                    rbga_mask = rgb_to_rgba(mask_image, ori_front_mask + ori_back_mask)
+                    mask_pil = Image.fromarray(cv2.cvtColor(rbga_mask.astype(np.uint8), cv2.COLOR_BGR2RGBA))
+                    image_data = io.BytesIO()
+                    mask_pil.save(image_data, format='PNG')
+                    image_data.seek(0)
+                    image_bytes = image_data.read()
+                    req = oss_upload_image(oss_client=self.minio_client, bucket="aida-clothing", object_name=f"mask/mask_{generate_uuid()}.png", image_bytes=image_bytes)
+                    result['mask_url'] = req.bucket_name + "/" + req.object_name
+
+
                     result_back_image = np.zeros_like(rgba_image)
                     back_mask = cv2.resize(back_mask, new_size, interpolation=cv2.INTER_AREA)
                     result_back_image[back_mask != 0] = rgba_image[back_mask != 0]

app/service/design_fast/utils/design_ensemble.py

@@ -10,12 +10,12 @@
 import logging
 
 import cv2
-import mmcv
 import numpy as np
 import torch
 import tritonclient.http as httpclient
 
 from app.core.config import DESIGN_MODEL_URL, DESIGN_MODEL_NAME
+from app.service.utils.image_normalize import my_imnormalize
 
 """
     keypoint
@@ -24,14 +24,14 @@ from app.core.config import DESIGN_MODEL_URL, DESIGN_MODEL_NAME
 
 
 def keypoint_preprocess(img_path):
-    img = mmcv.imread(img_path)
+    img = img_path
     img = cv2.copyMakeBorder(img, 25, 25, 25, 25, cv2.BORDER_CONSTANT, value=[255, 255, 255])
     img_scale = (256, 256)
     h, w = img.shape[:2]
     img = cv2.resize(img, img_scale)
     w_scale = img_scale[0] / w
     h_scale = img_scale[1] / h
-    img = mmcv.imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
+    img = my_imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
     preprocessed_img = np.expand_dims(img.transpose(2, 0, 1), axis=0)
     return preprocessed_img, (w_scale, h_scale)
 
@@ -78,7 +78,7 @@ def keypoint_postprocess(output, scale_factor):
 
 # KNet
 def seg_preprocess(img_path):
-    img = mmcv.imread(img_path)
+    img = img_path
     ori_shape = img.shape[:2]
     img_scale_w, img_scale_h = ori_shape
     if ori_shape[0] > 1024:
@@ -87,12 +87,12 @@ def seg_preprocess(img_path):
         img_scale_h = 1024
     # 如果图片size任意一边 大于 1024， 则会resize 成1024
     if ori_shape != (img_scale_w, img_scale_h):
-        # mmcv.imresize(img, img_scale_h, img_scale_w) # 老代码 引以为戒！哈哈哈~ h和w写反了
+        # my_imnormalize(img, img_scale_h, img_scale_w) # 老代码 引以为戒！哈哈哈~ h和w写反了
         img = cv2.resize(img, (img_scale_h, img_scale_w))
 
     # 扩充25的白边
     img = cv2.copyMakeBorder(img, 25, 25, 25, 25, cv2.BORDER_CONSTANT, value=[255, 255, 255])
-    img = mmcv.imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
+    img = my_imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
     preprocessed_img = np.expand_dims(img.transpose(2, 0, 1), axis=0)
     return preprocessed_img, ori_shape
 

app/service/design_fast/utils/organize.py

@@ -92,6 +92,8 @@ def organize_others(layer):
                        pattern_print_image_url=layer.get('pattern_print_image_url', None),
                        pattern_image=layer.get('pattern_image', None),
                        # back_perspective_url=layer['back_perspective_url'] if 'back_perspective_url' in layer.keys() else ""
+                       transpose=layer.get("transpose", [1, 1]),  # 默认为1, 1代表不镜像
+                       rotate=layer.get('rotate', 0),
                        )
     # 后片数据
     back_layer = dict(priority=-layer.get("priority", 0) if layer.get("layer_order", False) else PRIORITY_DICT.get(f'{layer["name"].lower()}_back', None),
@@ -109,6 +111,8 @@ def organize_others(layer):
                       pattern_overall_image_url=layer.get('pattern_overall_image_url', None),
                       pattern_print_image_url=layer.get('pattern_print_image_url', None),
                       # back_perspective_url=layer['back_perspective_url'] if 'back_perspective_url' in layer.keys() else ""
+                      transpose=layer.get("transpose", [1, 1]),  # 默认为1, 1代表不镜像
+                      rotate=layer.get('rotate', 0),
                       )
     return front_layer, back_layer
 

app/service/design_fast/utils/synthesis_item.py

@@ -347,7 +347,8 @@ def transpose_rotate(layer, image):
 
     rotate = layer.get('rotate', 0)
     paste_x, paste_y = layer['adaptive_position'][1], layer['adaptive_position'][0]
-
+    original_w = image.width
+    original_h = image.height
     # transpose左右是1 上下是-1
     if transpose[0] != 1:
         # 左右
@@ -361,8 +362,8 @@ def transpose_rotate(layer, image):
         image = image.rotate(-rotate, expand=True)
         # 4. 计算粘贴位置以保持视觉中心一致
         # 原本 (15, 36) 是 288*288 的左上角，我们计算其中心点
-        target_center_x = 15 + 288 // 2
+        target_center_x = paste_x + original_w // 2
-        target_center_y = 36 + 288 // 2
+        target_center_y = paste_y + original_h // 2
 
         # 获取旋转后图像的新尺寸
         new_w, new_h = image.size

app/service/design_pre_processing/service.py

@@ -7,7 +7,7 @@ import numpy as np
 import torch
 import tritonclient.grpc as grpcclient
 from minio import Minio
-from pymilvus import MilvusClient
+# from pymilvus import MilvusClient
 from urllib3.exceptions import ResponseError
 
 from app.core.config import settings, SR_MODEL_NAME, SR_TRITON_URL, MILVUS_TABLE_KEYPOINT, KEYPOINT_RESULT_TABLE_FIELD_SET
@@ -58,7 +58,21 @@ class DesignPreprocessing:
             if len(image.shape) == 2:
                 image = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)
             elif image.shape[2] == 4:  # 如果是四通道 mask
-                image = image[:, :, :3]
+                # 分离RGB和Alpha通道
+                bgr = image[:, :, :3]
+                alpha = image[:, :, 3]
+
+                # 创建白色背景（也可改为其他颜色，如(255,255,255)就是白色）
+                background_color = (255, 255, 255)
+                background = np.full_like(bgr, background_color)
+
+                # 将Alpha通道转换为掩码（0=透明，255=不透明）
+                alpha_mask = alpha / 255.0  # 归一化到0-1
+                alpha_mask = np.expand_dims(alpha_mask, axis=-1)  # 扩展维度，方便广播计算
+
+                # 混合背景和原图：透明区域显示背景色，不透明区域显示原图
+                image = (bgr * alpha_mask + background * (1 - alpha_mask)).astype(np.uint8)
+                # 此时image已经是3通道RGB，无需再执行image = image[:, :, :3]
             obj["image_obj"] = image
         return image_list
 
@@ -174,8 +188,9 @@ class DesignPreprocessing:
             scale = 0.4
             if waist_width / scale >= image_width:
                 add_width = int((waist_width / scale - image_width) / 2)
-                ret = cv2.copyMakeBorder(image['obj'], 0, 0, add_width, add_width, cv2.BORDER_CONSTANT, value=(256, 256, 256))
+                ret = cv2.copyMakeBorder(image['obj'], 0, 0, add_width, add_width, cv2.BORDER_CONSTANT, value=(255, 255, 255))
-                image_bytes = cv2.imencode(".jpg", ret)[1].tobytes()
+                img_rgba = cv2.cvtColor(ret, cv2.COLOR_RGB2RGBA)
+                image_bytes = cv2.imencode(".png", img_rgba)[1].tobytes()
                 # image['show_image_url'] = f"{image['image_url'].split('/', 1)[0]}/{self.minio_client.put_object(image['image_url'].split('/', 1)[0], image['image_url'].split('/', 1)[1].replace('.', '-show.'), io.BytesIO(image_bytes), len(image_bytes), content_type='image/jpeg').object_name}"
                 bucket_name = image['image_url'].split('/', 1)[0]
                 object_name = image['image_url'].split('/', 1)[1].replace('.', '-show.')
@@ -261,14 +276,15 @@ class DesignPreprocessing:
 
     def keypoint_cache(self, sketch):
         try:
-            client = MilvusClient(uri=settings.MILVUS_URL, token=settings.MILVUS_TOKEN, db_name=settings.MILVUS_ALIAS)
+            # client = MilvusClient(uri=settings.MILVUS_URL, token=settings.MILVUS_TOKEN, db_name=settings.MILVUS_ALIAS)
             keypoint_id = sketch['image_id']
-            res = client.query(
+            # res = client.query(
-                collection_name=MILVUS_TABLE_KEYPOINT,
+            #     collection_name=MILVUS_TABLE_KEYPOINT,
-                # ids=[keypoint_id],
+            #     # ids=[keypoint_id],
-                filter=f"keypoint_id == {keypoint_id}",
+            #     filter=f"keypoint_id == {keypoint_id}",
-                output_fields=['keypoint_vector', 'keypoint_site']
+            #     output_fields=['keypoint_vector', 'keypoint_site']
-            )
+            # )
+            res = []
             if len(res) == 0:
                 # 没有结果 直接推理拿结果 并保存
                 keypoint_infer_result = self.infer_keypoint_result(sketch)

app/service/generate_image/service_agent_tool_generate_image.py

@@ -11,7 +11,6 @@ import logging
 import uuid
 
 import cv2
-import mmcv
 import numpy as np
 import pandas as pd
 import torch
@@ -21,6 +20,7 @@ from minio import Minio
 from tritonclient.utils import np_to_triton_dtype
 
 from app.core.config import settings, FAST_GI_MODEL_URL, GI_MODEL_URL, DESIGN_MODEL_URL, FAST_GI_MODEL_NAME, GI_MODEL_NAME
+from app.service.utils.image_normalize import my_imnormalize
 from app.service.utils.new_oss_client import oss_upload_image
 
 logger = logging.getLogger()
@@ -86,10 +86,9 @@ class AgentToolGenerateImage:
 
     @staticmethod
     def preprocess(img):
-        img = mmcv.imread(img)
         img_scale = (224, 224)
         img = cv2.resize(img, img_scale)
-        img = mmcv.imnormalize(
+        img = my_imnormalize(
             img,
             mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]),
             to_rgb=True)

app/service/generate_image/service_generate_image.py

@@ -189,10 +189,10 @@ if __name__ == '__main__':
         tasks_id="123-89",
         prompt="a single item of sketch of dress, 4k, white background",
         image_url="aida-collection-element/89/Sketchboard/95f20cdc-e059-435c-b8b1-d04cc9e80c3d.png",
-        mode='img2img',
+        mode='txt2img',
         category="sketch",
         gender="Female",
-        version="fast"
+        version="hight"
     )
     server = GenerateImage(rd)
     print(server.get_result())

app/service/generate_image/utils/image_processing.py

@@ -2,23 +2,23 @@ import logging
 import time
 
 import cv2
-import mmcv
 import numpy as np
 import torch
 import tritonclient.http as httpclient
 
 from app.core.config import settings, DESIGN_MODEL_URL, DESIGN_MODEL_NAME
 from app.service.generate_image.utils.upload_sd_image import upload_stain_png_sd, upload_face_png_sd
+from app.service.utils.image_normalize import my_imnormalize
 
 logger = logging.getLogger()
 
 
 def seg_preprocess(img_path):
-    img = mmcv.imread(img_path)
+    img = img_path
     ori_shape = img.shape[:2]
     img_scale = ori_shape
     img = cv2.resize(img, img_scale)
-    img = mmcv.imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
+    img = my_imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
     preprocessed_img = np.expand_dims(img.transpose(2, 0, 1), axis=0)
     return preprocessed_img, ori_shape
 
@@ -242,10 +242,9 @@ def stain_detection(image, user_id, category, tasks_id, spot_size=100):
 
 def generate_category_recognition(image, gender):
     def preprocess(img):
-        img = mmcv.imread(img)
         img_scale = (224, 224)
         img = cv2.resize(img, img_scale)
-        img = mmcv.imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
+        img = my_imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
         preprocessed_img = np.expand_dims(img.transpose(2, 0, 1), axis=0)
         return preprocessed_img
 

app/service/lineart/service.py

@@ -1,7 +1,6 @@
 import logging
 
 import cv2
-import mmcv
 import numpy as np
 import torch
 import torch.nn.functional as F
@@ -10,6 +9,7 @@ from minio import Minio
 from app.core.config import settings
 from app.core.config import DESIGN_MODEL_URL
 from app.schemas.image2sketch import Image2SketchModel
+from app.service.utils.image_normalize import my_imnormalize
 from app.service.utils.new_oss_client import oss_get_image, oss_upload_image
 
 logger = logging.getLogger()
@@ -67,7 +67,7 @@ class LineArtService:
 
     @staticmethod
     def line_art_preprocess(image):
-        img = mmcv.imread(image)
+        img = image
         ori_shape = img.shape[:2]
         img_scale_w, img_scale_h = ori_shape
         if ori_shape[0] > 1024:
@@ -76,9 +76,9 @@ class LineArtService:
             img_scale_h = 1024
         # 如果图片size任意一边 大于 1024， 则会resize 成1024
         if ori_shape != (img_scale_w, img_scale_h):
-            # mmcv.imresize(img, img_scale_h, img_scale_w) # 老代码 引以为戒！哈哈哈~ h和w写反了
+            # my_imnormalize(img, img_scale_h, img_scale_w) # 老代码 引以为戒！哈哈哈~ h和w写反了
             img = cv2.resize(img, (img_scale_h, img_scale_w))
-        img = mmcv.imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
+        img = my_imnormalize(img, mean=np.array([123.675, 116.28, 103.53]), std=np.array([58.395, 57.12, 57.375]), to_rgb=True)
         preprocessed_img = np.expand_dims(img.transpose(2, 0, 1), axis=0)
         return preprocessed_img, ori_shape
 

app/service/prompt_generation/chatgpt_for_translation.py

@@ -90,7 +90,7 @@ def get_response(messages):
 
 def get_translation_from_llama3(text):
     start_time = time.time()
-    url = "http://10.1.1.240:11434/api/generate"
+    url = f"http://{settings.A6000_SERVICE_HOST}:12434/api/generate"
     # url = "http://10.1.1.240:1143/api/generate"
 
     # prompt = f"System: {prefix_for_llama}\nUser:[{text}]"
@@ -103,8 +103,8 @@ def get_translation_from_llama3(text):
 
     # 创建请求的负载 translator是自定义的翻译模型
     payload = {
-        "model": "translator",
+        "model": "AiDA-translator:latest",
-        "prompt": f"[{text}]",
+        "prompt": text,
         "stream": False
     }
     # 将负载转换为 JSON 格式
@@ -148,7 +148,7 @@ def get_translation_from_llama3(text):
 def get_prompt_from_image(image_path, text):
     start_time = time.time()
     # url = "http://localhost:11434/api/generate"
-    url = "http://10.1.1.243:11434/api/generate"
+    url = f"http://{settings.B_4_X_4090_SERVICE_HOST}:11434/api/generate"
 
     image_base64 = minio_util.minio_url_to_base64(image_path.img)
     # image_base64 = minio_url_to_base64(image_path)
@@ -180,7 +180,7 @@ def get_prompt_from_image(image_path, text):
 
 def main():
     """Main function"""
-    text = get_translation_from_llama3("[火焰]")
+    text = get_translation_from_llama3("火焰")
     print(text)
 
 

app/service/utils/image_normalize.py

@@ -0,0 +1,27 @@
+import cv2
+import numpy as np
+
+
+def my_imnormalize(img, mean, std, to_rgb=True):
+    """Inplace normalize an image with mean and std.
+
+    Args:
+        img (ndarray): Image to be normalized.
+        mean (ndarray): The mean to be used for normalize.
+        std (ndarray): The std to be used for normalize.
+        to_rgb (bool): Whether to convert to rgb.
+
+    Returns:
+        ndarray: The normalized image.
+    """
+    # cv2 inplace normalization does not accept uint8
+    img = img.copy().astype(np.float32)
+
+    assert img.dtype != np.uint8
+    mean = np.float64(mean.reshape(1, -1))
+    stdinv = 1 / np.float64(std.reshape(1, -1))
+    if to_rgb:
+        cv2.cvtColor(img, cv2.COLOR_BGR2RGB, img)  # inplace
+    cv2.subtract(img, mean, img)  # inplace
+    cv2.multiply(img, stdinv, img)  # inplace
+    return img

docker-compose.yml

@@ -1,25 +1,20 @@
 services:
   aida_server:
+    container_name: "AiDA_${SERVE_ENV}_Server"
     build:
       context: .
       dockerfile: Dockerfile
     working_dir: /app
     volumes:
       - ./app:/app/app
-      - ./.env_prod:/app/.env
+      - ./.env:/app/.env
       - /etc/localtime:/etc/localtime:ro
       - ./seg_cache:/seg_cache
     ports:
-      - "10200:80"
+      - "${SERVE_PORT}:80"
-    depends_on:
+    networks:
-      - redis
+      - aida_app_net
-  redis:
+networks:
-    image: redis
+  aida_app_net:
-    container_name: aida_redis
+    external: true
-    restart: always
+    name: aida_app_net
-    ports:
-      - "6400:6379"
-    volumes:
-      - ./redis/data:/data
-      - ./redis/conf/redis.conf:/etc/redis/redis.conf
-    command: redis-server /etc/redis/redis.conf --appendonly yes

pyproject.toml

@@ -23,8 +23,8 @@ dependencies = [
     "load-dotenv>=0.1.0",
     "loguru>=0.7.3",
     "minio>=7.2.20",
-    "mmcv>=2.2.0",
     "moviepy==1.0.3",
+    "np>=1.0.2",
     "numpy<2",
     "ollama>=0.6.1",
     "opencv-python>=4.11.0.86",

uv.lock

@@ -8,15 +8,6 @@ resolution-markers = [
     "(platform_machine != 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')",
 ]
 
-[[package]]
-name = "addict"
-version = "2.4.0"
-source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/85/ef/fd7649da8af11d93979831e8f1f8097e85e82d5bfeabc8c68b39175d8e75/addict-2.4.0.tar.gz", hash = "sha256:b3b2210e0e067a281f5646c8c5db92e99b7231ea8b0eb5f74dbdf9e259d4e494", size = 9186, upload-time = "2020-11-21T16:21:31.416Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/6a/00/b08f23b7d7e1e14ce01419a467b583edbb93c6cdb8654e54a9cc579cd61f/addict-2.4.0-py3-none-any.whl", hash = "sha256:249bb56bbfd3cdc2a004ea0ff4c2b6ddc84d53bc2194761636eb314d5cfa5dfc", size = 3832, upload-time = "2020-11-21T16:21:29.588Z" },
-]
-
 [[package]]
 name = "agentaction"
 version = "0.1.7"
@@ -1671,43 +1662,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/3e/9a/b697530a882588a84db616580f2ba5d1d515c815e11c30d219145afeec87/minio-7.2.20-py3-none-any.whl", hash = "sha256:eb33dd2fb80e04c3726a76b13241c6be3c4c46f8d81e1d58e757786f6501897e", size = 93751, upload-time = "2025-11-27T00:37:13.993Z" },
 ]
 
-[[package]]
-name = "mmcv"
-version = "2.2.0"
-source = { registry = "https://pypi.org/simple" }
-dependencies = [
-    { name = "addict" },
-    { name = "mmengine" },
-    { name = "numpy" },
-    { name = "opencv-python" },
-    { name = "packaging" },
-    { name = "pillow" },
-    { name = "pyyaml" },
-    { name = "regex", marker = "sys_platform == 'win32'" },
-    { name = "yapf" },
-]
-sdist = { url = "https://files.pythonhosted.org/packages/e9/a2/57a733e7e84985a8a0e3101dfb8170fc9db92435c16afad253069ae3f9df/mmcv-2.2.0.tar.gz", hash = "sha256:ac479247e808d8802f89eadf04d4118de86bdfe81361ec5aed0cc1bf731c67c9", size = 479121, upload-time = "2024-04-24T14:24:28.064Z" }
-
-[[package]]
-name = "mmengine"
-version = "0.10.7"
-source = { registry = "https://pypi.org/simple" }
-dependencies = [
-    { name = "addict" },
-    { name = "matplotlib" },
-    { name = "numpy" },
-    { name = "opencv-python" },
-    { name = "pyyaml" },
-    { name = "regex", marker = "sys_platform == 'win32'" },
-    { name = "rich" },
-    { name = "termcolor" },
-    { name = "yapf" },
-]
-sdist = { url = "https://files.pythonhosted.org/packages/17/14/959360bbd8374e23fc1b720906999add16a3ac071a501636db12c5861ff5/mmengine-0.10.7.tar.gz", hash = "sha256:d20ffcc31127567e53dceff132612a87f0081de06cbb7ab2bdb7439125a69225", size = 378090, upload-time = "2025-03-04T12:23:09.568Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/98/8e/f98332248aad102511bea4ae19c0ddacd2f0a994f3ca4c82b7a369e0af8b/mmengine-0.10.7-py3-none-any.whl", hash = "sha256:262ac976a925562f78cd5fd14dd1bc9b680ed0aa81f0d85b723ef782f99c54ee", size = 452720, upload-time = "2025-03-04T12:23:06.339Z" },
-]
-
 [[package]]
 name = "mmh3"
 version = "5.2.0"
@@ -1801,6 +1755,12 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/9e/c9/b2622292ea83fbb4ec318f5b9ab867d0a28ab43c5717bb85b0a5f6b3b0a4/networkx-3.6.1-py3-none-any.whl", hash = "sha256:d47fbf302e7d9cbbb9e2555a0d267983d2aa476bac30e90dfbe5669bd57f3762", size = 2068504, upload-time = "2025-12-08T17:02:38.159Z" },
 ]
 
+[[package]]
+name = "np"
+version = "1.0.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/40/7d/749666e5a9976dcbc4d16d487bbe571efc6bbf4cdf3f4620c0ccc52b57ef/np-1.0.2.tar.gz", hash = "sha256:781265283f3823663ad8fb48741aae62abcf4c78bc19f908f8aa7c1d3eb132f8", size = 7419, upload-time = "2017-10-05T11:26:00.956Z" }
+
 [[package]]
 name = "numpy"
 version = "1.26.4"
@@ -2269,15 +2229,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/bc/96/aaa61ce33cc98421fb6088af2a03be4157b1e7e0e87087c888e2370a7f45/pillow-12.0.0-cp312-cp312-win_arm64.whl", hash = "sha256:7dfb439562f234f7d57b1ac6bc8fe7f838a4bd49c79230e0f6a1da93e82f1fad", size = 2436012, upload-time = "2025-10-15T18:22:23.621Z" },
 ]
 
-[[package]]
-name = "platformdirs"
-version = "4.5.1"
-source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/cf/86/0248f086a84f01b37aaec0fa567b397df1a119f73c16f6c7a9aac73ea309/platformdirs-4.5.1.tar.gz", hash = "sha256:61d5cdcc6065745cdd94f0f878977f8de9437be93de97c1c12f853c9c0cdcbda", size = 21715, upload-time = "2025-12-05T13:52:58.638Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/cb/28/3bfe2fa5a7b9c46fe7e13c97bda14c895fb10fa2ebf1d0abb90e0cea7ee1/platformdirs-4.5.1-py3-none-any.whl", hash = "sha256:d03afa3963c806a9bed9d5125c8f4cb2fdaf74a55ab60e5d59b3fde758104d31", size = 18731, upload-time = "2025-12-05T13:52:56.823Z" },
-]
-
 [[package]]
 name = "posthog"
 version = "5.4.0"
@@ -2746,17 +2697,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/2c/58/ca301544e1fa93ed4f80d724bf5b194f6e4b945841c5bfd555878eea9fcb/referencing-0.37.0-py3-none-any.whl", hash = "sha256:381329a9f99628c9069361716891d34ad94af76e461dcb0335825aecc7692231", size = 26766, upload-time = "2025-10-13T15:30:47.625Z" },
 ]
 
-[[package]]
-name = "regex"
-version = "2025.11.3"
-source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/cc/a9/546676f25e573a4cf00fe8e119b78a37b6a8fe2dc95cda877b30889c9c45/regex-2025.11.3.tar.gz", hash = "sha256:1fedc720f9bb2494ce31a58a1631f9c82df6a09b49c19517ea5cc280b4541e01", size = 414669, upload-time = "2025-11-03T21:34:22.089Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/59/9b/7c29be7903c318488983e7d97abcf8ebd3830e4c956c4c540005fcfb0462/regex-2025.11.3-cp312-cp312-win32.whl", hash = "sha256:3839967cf4dc4b985e1570fd8d91078f0c519f30491c60f9ac42a8db039be204", size = 266194, upload-time = "2025-11-03T21:31:51.53Z" },
-    { url = "https://files.pythonhosted.org/packages/1a/67/3b92df89f179d7c367be654ab5626ae311cb28f7d5c237b6bb976cd5fbbb/regex-2025.11.3-cp312-cp312-win_amd64.whl", hash = "sha256:e721d1b46e25c481dc5ded6f4b3f66c897c58d2e8cfdf77bbced84339108b0b9", size = 277069, upload-time = "2025-11-03T21:31:53.151Z" },
-    { url = "https://files.pythonhosted.org/packages/d7/55/85ba4c066fe5094d35b249c3ce8df0ba623cfd35afb22d6764f23a52a1c5/regex-2025.11.3-cp312-cp312-win_arm64.whl", hash = "sha256:64350685ff08b1d3a6fff33f45a9ca183dc1d58bbfe4981604e70ec9801bbc26", size = 270330, upload-time = "2025-11-03T21:31:54.514Z" },
-]
-
 [[package]]
 name = "requests"
 version = "2.32.5"
@@ -3224,8 +3164,8 @@ dependencies = [
     { name = "load-dotenv" },
     { name = "loguru" },
     { name = "minio" },
-    { name = "mmcv" },
     { name = "moviepy" },
+    { name = "np" },
     { name = "numpy" },
     { name = "ollama" },
     { name = "opencv-python" },
@@ -3275,8 +3215,8 @@ requires-dist = [
     { name = "load-dotenv", specifier = ">=0.1.0" },
     { name = "loguru", specifier = ">=0.7.3" },
     { name = "minio", specifier = ">=7.2.20" },
-    { name = "mmcv", specifier = ">=2.2.0" },
     { name = "moviepy", specifier = "==1.0.3" },
+    { name = "np", specifier = ">=1.0.2" },
     { name = "numpy", specifier = "<2" },
     { name = "ollama", specifier = ">=0.6.1" },
     { name = "opencv-python", specifier = ">=4.11.0.86" },
@@ -3605,18 +3545,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/54/85/6ec269b0952ec7e36ba019125982cf11d91256a778c7c3f98a4c5043d283/xxhash-3.6.0-cp312-cp312-win_arm64.whl", hash = "sha256:eae5c13f3bc455a3bbb68bdc513912dc7356de7e2280363ea235f71f54064829", size = 27876, upload-time = "2025-10-02T14:34:54.371Z" },
 ]
 
-[[package]]
-name = "yapf"
-version = "0.43.0"
-source = { registry = "https://pypi.org/simple" }
-dependencies = [
-    { name = "platformdirs" },
-]
-sdist = { url = "https://files.pythonhosted.org/packages/23/97/b6f296d1e9cc1ec25c7604178b48532fa5901f721bcf1b8d8148b13e5588/yapf-0.43.0.tar.gz", hash = "sha256:00d3aa24bfedff9420b2e0d5d9f5ab6d9d4268e72afbf59bb3fa542781d5218e", size = 254907, upload-time = "2024-11-14T00:11:41.584Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/37/81/6acd6601f61e31cfb8729d3da6d5df966f80f374b78eff83760714487338/yapf-0.43.0-py3-none-any.whl", hash = "sha256:224faffbc39c428cb095818cf6ef5511fdab6f7430a10783fdfb292ccf2852ca", size = 256158, upload-time = "2024-11-14T00:11:39.37Z" },
-]
-
 [[package]]
 name = "yarl"
 version = "1.22.0"