生成对抗网

对抗样本与生成式对抗网络

生成对抗网络Generative Adversarial Nets

 生成式对抗网络GAN研究进展（一）

 生成式对抗网络GAN研究进展（二）——原始GAN

生成式对抗网络GAN研究进展（三）——条件GAN

 生成式对抗网络GAN研究进展（四）——Laplacian Pyramid of Adversarial Networks，LAPGAN

生成式对抗网络GAN研究进展（五）——Deep Convolutional Generative Adversarial Nerworks，DCGAN

对抗生成网络（Generative Adversarial Net)

DCGAN的小尝试（1）

在caffe 中实现Generative Adversarial Nets（一）

在caffe 中实现Generative Adversarial Nets（二）

GAN学习指南：从原理入门到制作生成Demo

手把手教你写一个生成对抗网络 — 生成对抗网络代码全解析, 详细代码解析(TensorFlow, numpy, matplotlib, scipy)

基于能量模型的生成对抗网络–生成对抗网络进阶

深度解读：GAN模型及其在2016年度的进展

近期GAN的模型和理论发展

http://lijiancheng0614.github.io/all-categories/

http://blog.evjang.com/2016/06/generative-adversarial-nets-in.html

Newmu/dcgan_code

goodfeli/adversarial

https://plus.google.com/+SoumithChintala/posts/MCtDVqsef6f

'''Deep Convolutional Generative Adversarial Network (DCGAN) TutorialThis tutorials walks through an implementation of DCGAN as described in Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks.To learn more about generative adversarial networks, see my Medium post on them.'''#Import the libraries we will need.import tensorflow as tfimport numpy as npimport input_dataimport matplotlib.pyplot as pltimport tensorflow.contrib.slim as slimimport osimport scipy.miscimport scipy'''We will be using the MNIST dataset. input_data is a library that downloads the dataset and uzips it automatically. It can be acquired Github here: https://gist.github.com/awjuliani/1d21151bc17362bf6738c3dc02f37906'''mnist = input_data.read_data_sets("MNIST_data/", one_hot=False)#Helper Functions#This function performns a leaky relu activation, which is needed for the discriminator network.def lrelu(x, leak=0.2, name="lrelu"):     with tf.variable_scope(name):         f1 = 0.5 * (1 + leak)         f2 = 0.5 * (1 - leak)         return f1 * x + f2 * abs(x)    #The below functions are taken from carpdem20's implementation https://github.com/carpedm20/DCGAN-tensorflow#They allow for saving sample images from the generator to follow progressdef save_images(images, size, image_path):    return imsave(inverse_transform(images), size, image_path)def imsave(images, size, path):    return scipy.misc.imsave(path, merge(images, size))def inverse_transform(images):    return (images+1.)/2.def merge(images, size):    h, w = images.shape[1], images.shape[2]    img = np.zeros((h * size[0], w * size[1]))    for idx, image in enumerate(images):        i = idx % size[1]        j = idx / size[1]        img[j*h:j*h+h, i*w:i*w+w] = image    return img'''Defining the Adversarial NetworksGenerator NetworkThe generator takes a vector of random numbers and transforms it into a 32x32 image. Each layer in the network involves a strided transpose convolution, batch normalization, and rectified nonlinearity. Tensorflow's slim library allows us to easily define each of these layers.'''def generator(z):        zP = slim.fully_connected(z,4*4*256,normalizer_fn=slim.batch_norm,\        activation_fn=tf.nn.relu,scope='g_project',weights_initializer=initializer)    zCon = tf.reshape(zP,[-1,4,4,256])        gen1 = slim.convolution2d_transpose(\        zCon,num_outputs=64,kernel_size=[5,5],stride=[2,2],\        padding="SAME",normalizer_fn=slim.batch_norm,\        activation_fn=tf.nn.relu,scope='g_conv1', weights_initializer=initializer)        gen2 = slim.convolution2d_transpose(\        gen1,num_outputs=32,kernel_size=[5,5],stride=[2,2],\        padding="SAME",normalizer_fn=slim.batch_norm,\        activation_fn=tf.nn.relu,scope='g_conv2', weights_initializer=initializer)        gen3 = slim.convolution2d_transpose(\        gen2,num_outputs=16,kernel_size=[5,5],stride=[2,2],\        padding="SAME",normalizer_fn=slim.batch_norm,\        activation_fn=tf.nn.relu,scope='g_conv3', weights_initializer=initializer)        g_out = slim.convolution2d_transpose(\        gen3,num_outputs=1,kernel_size=[32,32],padding="SAME",\        biases_initializer=None,activation_fn=tf.nn.tanh,\        scope='g_out', weights_initializer=initializer)        return g_out'''Discriminator NetworkThe discriminator network takes as input a 32x32 image and transforms it into a single valued probability of being generated from real-world data. Again we use tf.slim to define the convolutional layers, batch normalization, and weight initialization.'''def discriminator(bottom, reuse=False):        dis1 = slim.convolution2d(bottom,16,[4,4],stride=[2,2],padding="SAME",\        biases_initializer=None,activation_fn=lrelu,\        reuse=reuse,scope='d_conv1',weights_initializer=initializer)        dis2 = slim.convolution2d(dis1,32,[4,4],stride=[2,2],padding="SAME",\        normalizer_fn=slim.batch_norm,activation_fn=lrelu,\        reuse=reuse,scope='d_conv2', weights_initializer=initializer)        dis3 = slim.convolution2d(dis2,64,[4,4],stride=[2,2],padding="SAME",\        normalizer_fn=slim.batch_norm,activation_fn=lrelu,\        reuse=reuse,scope='d_conv3',weights_initializer=initializer)        d_out = slim.fully_connected(slim.flatten(dis3),1,activation_fn=tf.nn.sigmoid,\        reuse=reuse,scope='d_out', weights_initializer=initializer)        return d_out#Connecting them togethertf.reset_default_graph()z_size = 100 #Size of z vector used for generator.#This initializaer is used to initialize all the weights of the network.initializer = tf.truncated_normal_initializer(stddev=0.02)#These two placeholders are used for input into the generator and discriminator, respectively.z_in = tf.placeholder(shape=[None,z_size],dtype=tf.float32) #Random vectorreal_in = tf.placeholder(shape=[None,32,32,1],dtype=tf.float32) #Real imagesGz = generator(z_in) #Generates images from random z vectorsDx = discriminator(real_in) #Produces probabilities for real imagesDg = discriminator(Gz,reuse=True) #Produces probabilities for generator images#These functions together define the optimization objective of the GAN.d_loss = -tf.reduce_mean(tf.log(Dx) + tf.log(1.-Dg)) #This optimizes the discriminator.g_loss = -tf.reduce_mean(tf.log(Dg)) #This optimizes the generator.tvars = tf.trainable_variables()#The below code is responsible for applying gradient descent to update the GAN.trainerD = tf.train.AdamOptimizer(learning_rate=0.0002,beta1=0.5)trainerG = tf.train.AdamOptimizer(learning_rate=0.0002,beta1=0.5)d_grads = trainerD.compute_gradients(d_loss,tvars[9:]) #Only update the weights for the discriminator network.g_grads = trainerG.compute_gradients(g_loss,tvars[0:9]) #Only update the weights for the generator network.update_D = trainerD.apply_gradients(d_grads)update_G = trainerG.apply_gradients(g_grads)'''Training the networkNow that we have fully defined our network, it is time to train it!'''batch_size = 128 #Size of image batch to apply at each iteration.iterations = 500000 #Total number of iterations to use.sample_directory = './figs' #Directory to save sample images from generator in.model_directory = './models' #Directory to save trained model to.init = tf.initialize_all_variables()saver = tf.train.Saver()with tf.Session() as sess:      sess.run(init)    for i in range(iterations):        zs = np.random.uniform(-1.0,1.0,size=[batch_size,z_size]).astype(np.float32) #Generate a random z batch        xs,_ = mnist.train.next_batch(batch_size) #Draw a sample batch from MNIST dataset.        xs = (np.reshape(xs,[batch_size,28,28,1]) - 0.5) * 2.0 #Transform it to be between -1 and 1        xs = np.lib.pad(xs, ((0,0),(2,2),(2,2),(0,0)),'constant', constant_values=(-1, -1)) #Pad the images so the are 32x32        _,dLoss = sess.run([update_D,d_loss],feed_dict={z_in:zs,real_in:xs}) #Update the discriminator        _,gLoss = sess.run([update_G,g_loss],feed_dict={z_in:zs}) #Update the generator, twice for good measure.        _,gLoss = sess.run([update_G,g_loss],feed_dict={z_in:zs})        if i % 10 == 0:            print "Gen Loss: " + str(gLoss) + " Disc Loss: " + str(dLoss)            z2 = np.random.uniform(-1.0,1.0,size=[batch_size,z_size]).astype(np.float32) #Generate another z batch            newZ = sess.run(Gz,feed_dict={z_in:z2}) #Use new z to get sample images from generator.            if not os.path.exists(sample_directory):                os.makedirs(sample_directory)            #Save sample generator images for viewing training progress.            save_images(np.reshape(newZ[0:36],[36,32,32]),[6,6],sample_directory+'/fig'+str(i)+'.png')        if i % 1000 == 0 && i != 0:            if not os.path.exists(model_directory):                os.makedirs(model_directory)            saver.save(sess,model_directory+'/model-'+str(i)+'.cptk')            print "Saved Model"'''Using a trained networkOnce we have a trained model saved, we may want to use it to generate new images, and explore the representation it has learned.'''sample_directory = './figs' #Directory to save sample images from generator in.model_directory = './models' #Directory to load trained model from.batch_size_sample = 36init = tf.initialize_all_variables()saver = tf.train.Saver()with tf.Session() as sess:      sess.run(init)    #Reload the model.    print 'Loading Model...'    ckpt = tf.train.get_checkpoint_state(path)    saver.restore(sess,ckpt.model_checkpoint_path)        zs = np.random.uniform(-1.0,1.0,size=[batch_size_sample,z_size]).astype(np.float32) #Generate a random z batch    newZ = sess.run(Gz,feed_dict={z_in:z2}) #Use new z to get sample images from generator.    if not os.path.exists(sample_directory):        os.makedirs(sample_directory)    save_images(np.reshape(newZ[0:batch_size_sample],[36,32,32]),[6,6],sample_directory+'/fig'+str(i)+'.png')