使用百度AI开放平台处理数据集

亦凉 2022-12-28 13:00 238阅读 0赞

手里一堆人脸数据集需要处理，总不能手动，太费事了

于是使用一下百度的AI开放平台，地址如下：

[https://cloud.baidu.com/product/face][https_cloud.baidu.com_product_face]

使用起来非常简单，首先需要平台创建应用，此时就会有AK和SK

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3pob3VfNDM4_size_16_color_FFFFFF_t_70][]

这个用于获取token，有了token就可以直接请求了

获取token的代码：

注意填上做自己的AK和SK

# encoding:utf-8
    import requests 
    
    # client_id 为官网获取的AK， client_secret 为官网获取的SK
    host = 'https://aip.baidubce.com/oauth/2.0/token?grant_type=client_credentials&client_id=XXXXXX&client_secret=XXXXXXX'
    response = requests.get(host)
    if response:
        print(response.json())

token是一长串字符串，复制保存好

下面 我 实现和这个一个需求，获取人脸的性别和是否戴眼镜，然后分为四组，分别将文件需要移动的文件移动到这个四个文件夹中，想实现自己的需求自己根据这个可以进行修改

import os
    from shutil import move
    import time
    
    token="XXXXXXXXXXXXXXXXX"
    _url = "https://aip.baidubce.com/rest/2.0/face/v3/detect" + "?access_token=" + token
    headers = {'content-type': 'application/json'}
    trs='A'
    tr_ts=['train','test']
    path='F:\\XXX\\'+tr_ts[0]+trs#trainA,trainB,testA,testB
    ls=os.listdir(path)
    ls.sort()
    
    for i in ls:
        start=time.time()
        imgBase64=base64.b64encode(open(path+'\\'+i,'rb').read())
        data={"image": imgBase64, "image_type": "BASE64","face_field":"age,gender,glasses"}
        res= requests.post(_url,data=data,headers=headers).json()['result']['face_list'][0]
        #print(res)
        if (res['glasses']['type']!='none')&(res['gender']['type']=='male'):
            move(path+'\\'+i,trs+'\\man_glasses\\')
            print(time.time()-start)
        elif  (res['glasses']['type']!='none')&(res['gender']['type']=='female'):
            move(path+'\\'+i,trs+'\\woman_glasses\\')
            print(time.time()-start)
        elif (res['glasses']['type']=='none')&(res['gender']['type']=='male'):
            move(path+'\\'+i,trs+'\\man_noglasses\\')
            print(time.time()-start)
        else:move(path+'\\'+i,trs+'\\woman_noglasses\\');print(time.time()-start)

具体可以查看[这个链接][Link 1]因为识别出来的内容很多，我这里只设置了这些，在face\_field这个字段里面可以修改，可以看到我只需要age,gender,glasses

截取文档发的部分内容，可以看到非常多的属性，需要哪一个自己选择即可

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3pob3VfNDM4_size_16_color_FFFFFF_t_70 1][]

不过，按照上面的代码速度很慢，因为免费的就每秒可以2次，而且网速会限制

于是有制作了多进程版：

# encoding:utf-8
    import requests 
    import base64
    import os
    import time
    import multiprocessing as mp
    
    class MSP():
        def __init__(self):
            self.h='trainB'
            self.token="XXXXX"
            self._url = "https://aip.baidubce.com/rest/2.0/face/v3/detect" + "?access_token=" + self.token
            self.headers = {'content-type': 'application/json'}
            self.path='F:\\XXX\\'+self.h
            self.ls=os.listdir(self.path)
            self.ls.sort()
            self.lres=[]
            self.manager = mp.Manager
            self.mp_lst = self.manager().list()
        def post_func(self, i):
            self.imgBase64=base64.b64encode(open(self.path+'\\'+i,'rb').read())
            self.data={"image": self.imgBase64, "image_type": "BASE64","face_field":"age,gender,glasses"}
            self.res= requests.post(self._url,data=self.data,headers=self.headers).json()['result']['face_list'][0]
            self.res['name']=i
            self.mp_lst.append(self.res)
            time.sleep(0.1)
            print(i)
    
        def flow(self):
            pool = mp.Pool(10)
            for i in self.ls:
                pool.apply_async(self.post_func, args=(i,))            
            pool.close()
            pool.join()
    
    if __name__ == '__main__':
        start_time = time.time()
        msp = MSP()
        msp.flow()
        f=open('XXX.txt','w')
        f.write(str(msp.mp_lst))
        f.close()
        print(time.time() - start_time)

速度比之前快十倍左右，真的很快

将属性内容保存在一个txt，需要移动自己下代码读取文件内容即可

# encoding:utf-8
    import os
    from shutil import move
    import time
     
    #trainB
    path='F:\\XXX\\trainB'
    trs='B'
    lres_trainB=open('XXXX.txt').read()
    for res in eval(lres_trainB):
        start=time.time()
        i=res['name']
        if (res['glasses']['type']!='none')&(res['gender']['type']=='male'):
            move(path+'\\'+i,trs+'\\man_glasses\\')
            print(time.time()-start)
        elif  (res['glasses']['type']!='none')&(res['gender']['type']=='female'):
            move(path+'\\'+i,trs+'\\woman_glasses\\')
            print(time.time()-start)
        elif (res['glasses']['type']=='none')&(res['gender']['type']=='male'):
            move(path+'\\'+i,trs+'\\man_noglasses\\')
            print(time.time()-start)
        else:move(path+'\\'+i,trs+'\\woman_noglasses\\');print(time.time()-start)

至此，这个过程结束了

[https_cloud.baidu.com_product_face]: https://cloud.baidu.com/product/face
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3pob3VfNDM4_size_16_color_FFFFFF_t_70]: /images/20221120/9ca6a21f3bd84faab2b56a09b124aa05.png
[Link 1]: https://ai.baidu.com/ai-doc/FACE/yk37c1u4t
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3pob3VfNDM4_size_16_color_FFFFFF_t_70 1]: /images/20221120/f32ee148f0b14020a4847a57389bc134.png