毕业季--学士帽生成器（PaddleHub实现）

效果展示

原图：

Artbreeder

创建令人惊叹的插画和艺术

下载

变换后：

原图+学士帽：

（PS:示例图片均来源于互联网，如有侵权，请联系删除）

1、导入相关的包

import numpy as npimport matplotlib.image as mpimgimport matplotlib.pyplot as pltimport osfrom PIL import Imageimport paddlehub as hubimport cv2import shutil

2、展示需要人物图片

test_img_path = ["work/test.png"]#需要戴头饰的人物照片import matplotlib.pyplot as plt 
import matplotlib.image as mpimg 

img = mpimg.imread(test_img_path[0]) 

# 展示待预测图片plt.figure(figsize=(10,10))
plt.imshow(img) 
plt.axis('off') 
plt.show()

3、脸部检测

!pip install --upgrade paddlehub -i https://mirror.baidu.com/pypi/simple

import paddlehub as hub

module = hub.Module(name="ultra_light_fast_generic_face_detector_1mb_640")

input_dict = {"image": test_img_path}# execute predict and print the resultresults = module.face_detection(data=input_dict, visualization=True)for result in results:    print(result)

[2022-06-15 18:29:31,819] [ WARNING] - The _initialize method in HubModule will soon be deprecated, you can use the __init__() to handle the initialization of the object

{'data': [{'left': 322.3936462402344, 'right': 423.6573791503906, 'top': 264.9285583496094, 'bottom': 397.74920654296875, 'confidence': 0.9989694356918335}], 'path': 'work/test.png', 'save_path': 'face_detector_640_predict_output/test.png'}

4、合成带学士帽图片

可以对学士帽进行调节尺寸大小、角度

if os.path.exists('./HeCheng'):    pass      else:
    os.mkdir('./HeCheng')#创建文件夹test_path='work/test.png'#需要合成人物的图片toushi='work/学士帽.png'#学士帽照片#后处理：调整帽子maozi = Image.open('work/学士帽.png') 
maozi2copy  = maozi.copy() 
maozi2copy = maozi2copy.rotate(-5)#调整角度maozi2copy   = maozi2copy.resize((240,130))#调整大小savepath='HeCheng'#保存合成后图片文件夹x = results[0]if x.get('data'): 
    for result in results:        print(result)    # box 为头像在图片中的位置
    # box(x1, y1, x2, y2)
    # x1,y1 为头像左上角的位置
    # x2,y2 为头像右下角的位置
    x1 = results[0]['data'][0]['left']
    y1 = results[0]['data'][0]['top']
    x2 = results[0]['data'][0]['right']
    y2 = results[0]['data'][0]['bottom']    
    
    # 学士帽图片尺寸，和resize相同
    fruit_size = (240, 130)    
    
    #调整偏移量
    addx = -65 #这个可以根据实际情况调节,越大越往右（注意是负值）
    addy = -50 #这个可以根据实际情况调节，越大越往下（注意是负值）
    box = (int(x1)+addx,int(y1)+addy,int(x1)+fruit_size[0]+addx,int(y1)+ fruit_size[1]+addy)    #print(x1)

    pil_im1 = Image.open(test_path)    # pil_im2 = Image.open(toushi).convert('RGBA')
    region = pil_im1.crop(box)#cut from the picture
    region = region.transpose(Image.ROTATE_270)#rotate the image
    pil_im1.paste(maozi2copy, box, maozi2copy)
    plt.imshow(pil_im1)
    pil_im1.save(savepath+'/'+'合成.png')#合成后的图片else:    print("没有检测到")

{'data': [{'left': 322.3936462402344, 'right': 423.6573791503906, 'top': 264.9285583496094, 'bottom': 397.74920654296875, 'confidence': 0.9989694356918335}], 'path': 'work/test.png', 'save_path': 'face_detector_640_predict_output/test.png'}

5、HubServing轻量级服务化部署

!hub serving start -m ultra_light_fast_generic_face_detector_1mb_640

这一步请在 终端-1 执行 执行成功后会出现以下画面

# !hub serving start -m ultra_light_fast_generic_face_detector_1mb_640

然后通过客服端向服务器发送请求

发送成功后会出现以下画面

# coding: utf8%matplotlib inlineimport matplotlib.pyplot as plt # plt 用于显示图片import requestsimport jsonimport cv2import base64import numpy as npimport colorsysimport warningsimport matplotlib.image as mpimgimport osfrom PIL import Imageimport paddlehub as hubimport shutil

warnings.filterwarnings("ignore")
plt.figure(figsize=(12,8))def cv2_to_base64(image):
    data = cv2.imencode('.jpg', image)[1]    return base64.b64encode(data.tostring()).decode('utf8')if __name__ == '__main__':    # 获取图片的base64编码格式
    img1 = cv2_to_base64(cv2.imread("/home/aistudio/work/test.png"))    #img2 = cv2_to_base64(cv2.imread("MyDataset/images/hard_hat_workers1457.png"))
    data = {'images': [img1]}    # data = {'images': [img1]}
    # 指定content-type
    headers = {"Content-type": "application/json"}    # 发送HTTP请求
    url = "http://127.0.0.1:8866/predict/ultra_light_fast_generic_face_detector_1mb_640"
    r = requests.post(url=url, headers=headers, data=json.dumps(data))    # 打印预测结果，注意，r.json()["results"]本身就是一个数组，要取到对应图片的人脸关键点预测结果
    print(r.json()["results"])    
    if os.path.exists('./HeCheng'):        pass      
    else:
        os.mkdir('./HeCheng')#创建文件夹


    test_path='/home/aistudio/work/test.png'#需要合成人物的图片
    toushi='/home/aistudio/work/学士帽.png'#学士帽照片

    #后处理：调整帽子
    maozi = Image.open('work/学士帽.png') 
    maozi2copy  = maozi.copy() 
    maozi2copy = maozi2copy.rotate(-5)#调整角度
    maozi2copy   = maozi2copy.resize((240,130))#调整大小


    savepath='HeCheng'#保存合成后图片文件夹
    x = results[0]    if x.get('data'): 
        for result in results:            print(result)        # box 为头像在图片中的位置
        # box(x1, y1, x2, y2)
        # x1,y1 为头像左上角的位置
        # x2,y2 为头像右下角的位置
        x1 = results[0]['data'][0]['left']
        y1 = results[0]['data'][0]['top']
        x2 = results[0]['data'][0]['right']
        y2 = results[0]['data'][0]['bottom']        
        
        # 学士帽图片尺寸，和resize相同
        fruit_size = (240, 130)        
        
        #调整偏移量
        addx = -65 #这个可以根据实际情况调节,越大越往右（注意是负值）
        addy = -50 #这个可以根据实际情况调节，越大越往下（注意是负值）
        box = (int(x1)+addx,int(y1)+addy,int(x1)+fruit_size[0]+addx,int(y1)+ fruit_size[1]+addy)        #print(x1)

        pil_im1 = Image.open(test_path)        # pil_im2 = Image.open(toushi).convert('RGBA')
        region = pil_im1.crop(box)#cut from the picture
        region = region.transpose(Image.ROTATE_270)#rotate the image
        pil_im1.paste(maozi2copy, box, maozi2copy)
        plt.imshow(pil_im1)
        pil_im1.save(savepath+'/'+'合成.png')#合成后的图片
    else:        print("没有检测到")

[{'data': [{'bottom': 397.8898620605469, 'confidence': 0.9991759657859802, 'left': 321.9490051269531, 'right': 423.74810791015625, 'top': 265.1085510253906}]}]
{'data': [{'left': 322.3936462402344, 'right': 423.6573791503906, 'top': 264.9285583496094, 'bottom': 397.74920654296875, 'confidence': 0.9989694356918335}], 'path': 'work/test.png', 'save_path': 'face_detector_640_predict_output/test.png'}