以图搜图(demo创建流程)

window10添加向量数据库以及调用

 

 

创建docker

1,在windows功能中打开Hyper-V 和 容器

2,进入https://www.docker.com/ ,下载windows版本进行安装

创建milvus及连接

1,创建milvus文件夹,在文件夹下建立confdblogspicwal五个文件夹,把docker-compose.ymlserver_config.yaml放如conf文件夹中,使用命令提示符进入到conf文件夹中,执行docker-compose up -d,执行结束后,再执行docker-compose up -d来查看运行状态,然后运行docker run -p 3000:3000 -e HOST_URL=http://192.168.10.171:3000 -e MILVUS_URL=192.168.10.171:19530 milvusdb/milvus-insight:latest 。在docker中即可看到

 

 

点击3000端口,通过3000接口进入到milvus界面。

 

2,python程序

1》创建集合

from pymilvus import FieldSchema, DataType, CollectionSchema, connections, Collection

#创建存储字段

id = FieldSchema(

    name="id",

    dtype=DataType.INT64,

    is_primary=True,

)

vec = FieldSchema(

    name="vec",

    dtype=DataType.FLOAT_VECTOR,

    dim=2048

)

schema = CollectionSchema(

    fields=[id, vec],

    description="Test search"

)

collection_name = "animal"

 

print("#########连接数据库##############")

connections.connect(

    alias="default",

    host='192.168.10.171',

    port='19530'

)

 

print("#########根据上面得信息创建集合##############")

collection = Collection(

    name=collection_name,

    schema=schema,

    using='default',

    shards_num=2

)

 

print("#########关闭连接##############")

connections.disconnect("default")

 

2》添加数据

#print("#########连接数据库##############")

from towhee import pipeline

from pymilvus import connections, Collection

 

connections.connect(

    alias="default",

    host='192.168.10.171',

    port='19530'

)

#print("#########插入数据##############")

 

p = pipeline('image-embedding')

output = p('http://localhost:54867/7.jpg')

 

data = [

    [7],

    [output],

]

 

collection = Collection("animal")  

mr = collection.insert(data)

print(mr)

 

3》查询数据

import towhee

from django.http import HttpResponse

from pymilvus import connections,Collection

from towhee import pipeline

 

 

#获取图片向量查询的数据

def selectData(request):

    connections.connect(host='192.168.10.171', port='19530')

    t = (             

        towhee.glob['path']('E://milvus//coreCode//getImages//getImages//wwwroot//7.jpg')

        .image_decode['path', 'img']()

        .image_embedding.timm['img', 'vec'](model_name='resnet50')

        .milvus_search['vec', 'results'](collection='animal')

        .select['results']()  # 选择指定列;

        .to_list()

    )

    tr = "";

    for i in t:

        for j in i.results:

            tr = tr + "  " + str(j.id)

    return HttpResponse(tr, content_type="application/json")

 

def selectvector(request):

    #print("#########连接数据库##############")

    connections.connect(

        alias="default",

        host='192.168.10.171',

        port='19530'

    )

    #print("#########查询数据##############")

    p = pipeline('image-embedding')

    output = p('http://localhost:54867/7.jpg')

    collection = Collection("animal")  

 

    search_params = {"metric_type": "L2", "params": {"nprobe": 1}, "search_length": 100}

 

    tt = collection.search(

        data=[output],

        anns_field="vec",

        param=search_params,

        offset=0,

        limit=3,

        expr=None

    )

    #print("#########查询结果##############")

    return HttpResponse(tt, content_type="application/json")

posted @ 2023-01-31 16:06  zwbsoft  阅读(116)  评论(0编辑  收藏  举报