用Python十秒做表白神器！虽然520已经过去了，但是还有七夕啊！

概述<divstyle=\"margin:0px;padding:0px;border:0px;font-size:14px;line-height:inherit;font-family:helvetica,Arial,\'HiraginoSansGB\',\'MicrosoftYaHei\',simsun;vertical-align:baseline;color:rgb(89,8

<div ><div ><div ><div ><div ><div ><div ><p >520小编也是吃到了一大波狗粮啊，有钱的超级浪漫，没钱的也很会玩！所以小编今天决定还是教大家来做一款表白神器，就算这次用不着没下次也是肯定可以用的着的！

<p >

<p >

<p >

<p >今天，我就来教大家一下，<span >如何用Python做一份特别的礼物送给自己的恋人。

<p >

<p >当然了，如果还是单身的，也可以把这个作为<span >表白神器，和心爱的人<span >表白。

<p >会Python编程的人当然不用我说，就知道该如何 *** 作，<span >那些不懂编程的人，如果想尝试，那该怎么办呢？

<p >

<p >

<p >首先教大家一个初级版的。这个就比较简单，利用Python制作一个爱心。

<p >我先把代码给贴出来：

<p >import turtle

<p >import time

<p ># 画爱心的顶部

<p >def littleHeart():

<p >for i in range (200):

<p >turtle.right(1)

<p >turtle.forward(2)

<p ># 输入表白的语句，默认I love you

<p >love=input('Please enter a sentence of love,otherwise the default is "I love you":\n')

<p >#输入署名或者赠谁，没有不执行

<p >me=input('Please enter pen name,otherwise the default do not execute:\n')

<p >if love=='':

<p >love='I love you'

<p ># 窗口大小

<p >turtle.setup(wIDth=900,height=500)

<p ># 颜色

<p >turtle.color('red','pink')

<p ># 笔粗细

<p >turtle.pensize(3)

<p ># 速度

<p >turtle.speed(1)

<p ># 提笔

<p >turtle.up()

<p ># 隐藏笔

<p >turtle.hIDeturtle()

<p ># 去到的坐标,窗口中心为0,0

<p >turtle.goto(0,-180)

<p >turtle.showturtle()

<p ># 画上线

<p >turtle.down()

<p >turtle.speed(1)

<p >turtle.begin_fill()

<p >turtle.left(140)

<p >turtle.forward(224)

<p >#调用画爱心左边的顶部

<p >littleHeart()

<p >#调用画爱右边的顶部

<p >turtle.left(120)

<p >littleHeart()

<p ># 画下线

<p >turtle.forward(224)

<p >turtle.end_fill()

<p >turtle.pensize(5)

<p >turtle.up()

<p >turtle.hIDeturtle()

<p ># 在心中写字 一次

<p >turtle.goto(0,0)

<p >turtle.showturtle()

<p >turtle.color('#CD5C5C','pink')

<p >#在心中写字 Font可以设置字体自己电脑有的都可以设 align开始写字的位置

<p >turtle.write(love,Font=('gungsuh',30,),align="center")

<p >turtle.up()

<p >turtle.hIDeturtle()

<p >time.sleep(2)

<p ># 在心中写字 二次

<p >turtle.goto(0,0)

<p >turtle.showturtle()

<p >turtle.color('red','pink')

<p >turtle.write(love,align="center")

<p >turtle.up()

<p >turtle.hIDeturtle()

<p ># 写署名

<p >if me !='':

<p >turtle.color('black','pink')

<p >time.sleep(2)

<p >turtle.goto(180,-180)

<p >turtle.showturtle()

<p >turtle.write(me,Font=(20,align="center",move=True)

<p >#点击窗口关闭

<p >window=turtle.Screen()

<p >window.exitonclick()

<p >这个代码最终呈现效果如下，这个是比较初级简单的爱心，没有什么高难度。你也可以把代码扩充一下，整的更加高大上一些。

<p >

<p >

<p >import turtle

<p >import random

<p >def love(x,y): # 在(x,y)处画爱心lalala

<p >lv = turtle.Turtle()

<p >lv.hIDeturtle()

<p >lv.up()

<p >lv.goto(x,y) # 定位到(x,y)

<p >def curvemove(): # 画圆弧

<p >for i in range(20):

<p >lv.right(10)

<p >lv.forward(2)

<p >lv.color('red','pink')

<p >lv.speed(10000000)

<p >lv.pensize(1)

<p ># 开始画爱心lalala

<p >lv.down()

<p >lv.begin_fill()

<p >lv.left(140)

<p >lv.forward(22)

<p >curvemove()

<p >lv.left(120)

<p >curvemove()

<p >lv.forward(22)

<p >lv.write("WM",Font=("Arial",12,"normal"),align="center") # 写上表白的人的名字

<p >lv.left(140) # 画完复位

<p >lv.end_fill()

<p >def tree(branchLen,t):

<p >if branchLen > 5: # 剩余树枝太少要结束递归

<p >if branchLen < 20: # 如果树枝剩余长度较短则变绿

<p >t.color("green")

<p >t.pensize(random.uniform((branchLen + 5) / 4 - 2,(branchLen + 6) / 4 + 5))

<p >t.down()

<p >t.forward(branchLen)

<p >love(t.xcor(),t.ycor()) # 传输现在turtle的坐标

<p >t.up()

<p >t.backward(branchLen)

<p >t.color("brown")

<p >return

<p >t.pensize(random.uniform((branchLen + 5) / 4 - 2,(branchLen + 6) / 4 + 5))

<p >t.down()

<p >t.forward(branchLen)

<p ># 以下递归

<p >ang = random.uniform(15,45)

<p >t.right(ang)

<p >tree(branchLen - random.uniform(12,16),t) # 随机决定减小长度

<p >t.left(2 * ang)

<p >tree(branchLen - random.uniform(12,t) # 随机决定减小长度

<p >t.right(ang)

<p >t.up()

<p >t.backward(branchLen)

<p >myWin = turtle.Screen()

<p >t = turtle.Turtle()

<p >t.hIDeturtle()

<p >t.speed(1000)

<p >t.left(90)

<p >t.up()

<p >t.backward(200)

<p >t.down()

<p >t.color("brown")

<p >t.pensize(32)

<p >t.forward(60)

<p >tree(100,t)

<p >myWin.exitonclick()

<p >

<p >

<p >

<p >

<p ><span >先第一个：画像重叠。

<p >我们先选择两幅画，你们也可以一幅选你们心上人的画像，一幅选择风景或者其他。这个时候就看各位的审美了。这里我选择的都是风景。

<p >

<p >

<p >

<p >

<p >再获取图片宽高：

<p ># 获取图片的最小宽高

<p >wIDth = min(img1.size[0],img2.size[0])

<p >height = min(img1.size[1],img2.size[1])

<p >img_new = Image.new('RGB',(wIDth,height))

<p >这时候渲染图片：

<p ># 渲染图片

<p >for x in range(wIDth):

<p >for y in range(height):

<p >r1,g1,b1=img1.getpixel((x,y))

<p >r2,g2,b2=img2.getpixel((x,y))

<p >r=int(percent1r1+percent2r2)

<p >g=int(percent1g1+percent2g2)

<p >b=int(percent1b1+percent2b2)

<p >img_new.putpixel((x,y),(r,g,b))

<p >最后保存就好了！

<p ># 保存图片

<p >img_new.save('new.jpg')

<p >

<p ><span >第二个是图像渲染：

<p ><span >通过Python的深度学习算法包去训练计算机模仿世界名画的风格，然后应用到另一幅画中！

<p >这个就没有小程序了。因为这个有几百个依赖包。

<p >专业难度比较高一些，首先，需要安装使用的模块，pip一键搞定：

<p >pip3 install keras

<p >pip3 install h5py

<p >pip3 install tensorflow

<p >TensorFlow的安装可能不翻墙的话下载的比较慢，也可以源码安装。自己把握。（TensorFlow只能python3.5安装，所以先下载一个3.5版本的）

<p >然后再下载VGG16模型。把代码生成py格式和需要渲染图片放在同一个文件夹。

<p >

<p >我先把代码贴出来（这个代码是根据知乎大佬：杨航锋的代码修改而成）：

<p >from future import print_function

<p >from keras.preprocessing.image import load_img,img_to_array

<p >from scipy.misc import imsave

<p >import numpy as np

<p >import time

<p >import argparse

<p >from keras.applications import vgg16

<p >from keras import backend as K

<p >from scipy.optimize.lbfgsb import fmin_l_bfgs_b

<p >parser = argparse.ArgumentParser(description='Neural style transfer with Keras.')

<p >parser.add_argument('base_image_path',Metavar='base',type=str,help='Path to the image to transform.')

<p >parser.add_argument('style_reference_image_path',Metavar='ref',help='Path to the style reference image.')

<p >parser.add_argument('result_prefix',Metavar='res_prefix',help='Prefix for the saved results.')

<p >parser.add_argument('--iter',type=int,default=15,required=False,help='Number of iterations to run.')

<p >parser.add_argument('--content_weight',type=float,default=0.025,help='Content weight.')

<p >parser.add_argument('--style_weight',default=1.0,help='Style weight.')

<p >parser.add_argument('--tv_weight',help='Total Variation weight.')

<p >args = parser.parse_args()

<p >base_image_path = args.base_image_path

<p >style_reference_image_path = args.style_reference_image_path

<p >result_prefix = args.result_prefix

<p >iterations = args.iter

<p ># 不同损失分量的权重

<p >total_variation_weight = args.tv_weight

<p >style_weight = args.style_weight

<p >content_weight = args.content_weight

<p ># 生成图片的尺寸

<p >wIDth,height = load_img(base_image_path).size

<p >img_nrows = 400

<p >img_ncols = int(wIDth * img_nrows / height)

<p ># util function to open,调整和格式化图片到适当的张量

<p >def preprocess_image(image_path):

<p >img = load_img(image_path,target_size=(img_nrows,img_ncols))

<p >img = img_to_array(img)

<p >img = np.expand_dims(img,axis=0)

<p >img = vgg16.preprocess_input(img)

<p >return img

<p ># util函数将一个张量转换成一个有效的图像

<p >def deprocess_image(x):

<p >if K.image_data_format() == 'channels_first':

<p >x = x.reshape((3,img_nrows,img_ncols))

<p >x = x.transpose((1,2,0))

<p >else:

<p >x = x.reshape((img_nrows,img_ncols,3))

<p ># Remove zero-center by mean pixel

<p ># 用平均像素去除零中心

<p >x[:,:,0] += 103.939

<p >x[:,1] += 116.779

<p >x[:,2] += 123.68

@H313301@'RGB' 转换

<p >x = x[:,::-1]

<p >x = np.clip(x,255).astype('uint8')

<p >return x

<p ># get tensor representations of our images

<p ># 得到图像的张量表示

<p >base_image = K.variable(preprocess_image(base_image_path))

<p >style_reference_image = K.variable(preprocess_image(style_reference_image_path))

<p ># this will contain our generated image

<p ># 包含我们生成的图片

<p >if K.image_data_format() == 'channels_first':

<p >combination_image = K.placeholder((1,3,img_ncols))

<p >else:

<p >combination_image = K.placeholder((1,3))

<p ># combine the 3 images into a single Keras tensor

<p ># 将3个图像合并成一个Keras张量

<p >input_tensor = K.concatenate([base_image,

<p >style_reference_image,

<p >combination_image],axis=0)

<p ># build the VGG16 network with our 3 images as input

<p ># the model will be loaded with pre-trained ImageNet weights

<p ># 以我们的3个图像作为输入构建VGG16网络

<p ># 该模型将加载预先训练的ImageNet权重

<p >model = vgg16.VGG16(input_tensor=input_tensor,

<p >weights='imagenet',include_top=False)

<p >print('Model loaded.')

<p ># get the symbolic outputs of each "key" layer (we gave them unique names).

<p ># 获取每个“键”层的符号输出(我们给它们取了唯一的名称)

<p >outputs_dict = dict([(layer.name,layer.output) for layer in model.layers])

<p ># compute the neural style loss

<p ># 计算神经类型的损失

<p ># first we need to define 4 util functions

<p ># 首先我们需要定义是个until函数

<p ># the gram matrix of an image tensor (feature-wise outer product)

<p ># 图像张量的克矩阵

<p >def gram_matrix(x):

<p >assert K.ndim(x) == 3

<p >if K.image_data_format() == 'channels_first':

<p >features = K.batch_flatten(x)

<p >else:

<p >features = K.batch_flatten(K.permute_dimensions(x,(2,1)))

<p >gram = K.dot(features,K.transpose(features))

<p >return gram

<p ># the "style loss" is designed to maintain

<p ># 风格损失”是为了维护而设计的

<p ># the style of the reference image in the generated image.

<p ># 生成图像中引用图像的样式

<p ># It is based on the gram matrices (which capture style) of feature maps from the style reference image and from the generated image

<p ># 它基于从样式引用图像和生成的图像中获取特征映射的gram矩阵(捕获样式)

<p >def style_loss(style,combination):

<p >assert K.ndim(style) == 3

<p >assert K.ndim(combination) == 3

<p >S = gram_matrix(style)

<p >C = gram_matrix(combination)

<p >channels = 3

<p >size = img_nrows img_ncols

<p >return K.sum(K.square(S - C)) / (4. (channels * 2) (size ** 2))

<p ># an auxiliary loss function

<p ># 一个辅助的损失函数

<p ># designed to maintain the "content" of the base image in the generated image

<p >#设计用于维护生成图像中基本图像的“内容

<p >def content_loss(base,combination):

<p >return K.sum(K.square(combination - base))

<p ># the 3rd loss function,total variation loss,designed to keep the generated image locally coherent

<p ># 第三个损失函数，总变异损失，设计来保持生成的图像局部一致

<p >def total_variation_loss(x):

<p >assert K.ndim(x) == 4

<p >if K.image_data_format() == 'channels_first':

<p >a = K.square(x[:,:img_nrows - 1,:img_ncols - 1] - x[:,1:,:img_ncols - 1])

<p >b = K.square(x[:,1:])

<p >else:

<p >a = K.square(x[:,:img_ncols - 1,:] - x[:,:])

<p >b = K.square(x[:,:])

<p >return K.sum(K.pow(a + b,1.25))

<p ># combine these loss functions into a single scalar

<p ># 将这些损失函数合并成一个标量。

<p >loss = K.variable(0.)

<p >layer_features = outputs_dict['block4_conv2']

<p >base_image_features = layer_features[0,:]

<p >combination_features = layer_features[2,:]

<p >loss += content_weight * content_loss(base_image_features,

<p >combination_features)

<p >feature_layers = ['block1_conv1','block2_conv1',

<p >'block3_conv1','block4_conv1',

<p >'block5_conv1']

<p >for layer_name in feature_layers:

<p >layer_features = outputsdict[layername]

<p >style_reference_features = layer_features[1,:]

<p >sl = style_loss(style_reference_features,combination_features)

<p >loss += (style_weight / len(feature_layers)) sl

<p >loss += total_variation_weight total_variation_loss(combination_image)

<p ># get the gradIEnts of the generated image wrt the loss

<p ># 得到所生成图像的梯度，并对损失进行wrt。

<p >grads = K.gradIEnts(loss,combination_image)

<p >outputs = [loss]

<p >if isinstance(grads,(List,tuple)):

<p >outputs += grads

<p >else:

<p >outputs.append(grads)

<p >f_outputs = K.function([combination_image],outputs)

<p >def eval_loss_and_grads(x):

<p >if K.image_data_format() == 'channels_first':

<p >x = x.reshape((1,img_ncols))

<p >else:

<p >x = x.reshape((1,3))

<p >outs = f_outputs()

<p >loss_value = outs[0]

<p >if len(outs[1:]) == 1:

<p >grad_values = outs[1].flatten().astype('float64')

<p >else:

<p >grad_values = np.array(outs[1:]).flatten().astype('float64')

<p >return loss_value,grad_values

<p >"""

<p >this Evaluator class makes it possible

<p >to compute loss and gradIEnts in one pass

<p >while retrIEving them via two separate functions,

<p >"loss" and "grads". This is done because scipy.optimize

<p >requires separate functions for loss and gradIEnts,

<p >but computing them separately would be inefficIEnt.

<p >这个评估器类使它成为可能。

<p >在一个通道中计算损耗和梯度。

<p >当通过两个不同的函数检索它们时，

<p >“损失”和“梯度”。这是因为scipy.optimize

<p >要求分离的函数用于损失和梯度，

<p >但是单独计算它们将是低效的

<p >"""

<p >class Evaluator(object):

<p >def init(self):

<p >self.loss_value = None

<p >self.grads_values = None

<p >def loss(self,x):

<p >assert self.loss_value is None

<p >loss_value,grad_values = eval_loss_and_grads(x)

<p >self.loss_value = loss_value

<p >self.grad_values = grad_values

<p >return self.loss_value

<p >def grads(self,x):

<p >assert self.loss_value is not None

<p >grad_values = np.copy(self.grad_values)

<p >self.loss_value = None

<p >self.grad_values = None

<p >return grad_values

<p >evaluator = Evaluator()

<p ># run scipy-based optimization (L-BFGS) over the pixels of the generated image

<p ># 运行 scipy-based optimization (L-BFGS) 覆盖 生成的图像的像素

<p ># so as to minimize the neural style loss

<p ># 这样可以减少神经类型的损失

<p >if K.image_data_format() == 'channels_first':

<p >x = np.random.uniform(0,(1,img_ncols)) - 128.

<p >else:

<p >x = np.random.uniform(0,3)) - 128.

<p >for i in range(iterations):

<p >print('Start of iteration',i)

<p >start_time = time.time()

<p >x,min_val,info = fmin_l_bfgs_b(evaluator.loss,x.flatten(),

<p >fprime=evaluator.grads,maxfun=20)

<p >print('Current loss value:',min_val)

<p ># save current generated image

<p >img = deprocess_image(x.copy())

<p >fname = result_prefix + '_at_iteration_%d.png' % i

<p >imsave(fname,img)

<p >end_time = time.time()

<p >print('Image saved as',fname)

<p >print('Iteration %d completed in %ds' % (i,end_time - start_time))

<p >我先把代码贴出来（这个代码是根据知乎大佬：杨航锋的代码修改而成）：

<p >from future import print_function

<p >from keras.preprocessing.image import load_img,end_time - start_time))

<p >它会有一个不断渐进渲染的过程：

<p >

<p >虽然我有老婆，而且我老婆特别好看，漂亮。但是为了不伤害到你们，我就用万门的新起点嘉园大楼渲染一下莫奈的名画。给你们具体看一下。

<p >

<p >

<p >

<p >

<p >

<p >

<p >

<p >其实，只要是自己用心做出的礼物，你喜欢的人一定会非常感动。

<p >

<p >愿每一个渴望恋爱的人都能在520这天找到自己的心有所属。

<p >

<p >转载于：万门

<p >欢迎大家关注我的博客或者公众号：https://home.cnblogs.com/u/Python1234/ Python学习交流

总结

以上是内存溢出为你收集整理的用Python十秒做表白神器！虽然520已经过去了，但是还有七夕啊！全部内容，希望文章能够帮你解决用Python十秒做表白神器！虽然520已经过去了，但是还有七夕啊！所遇到的程序开发问题。

如果觉得内存溢出网站内容还不错，欢迎将内存溢出网站推荐给程序员好友。