WGAN-GP生成MNIST
参考博客点击打开链接
33个epoch结果
#coding:utf-8 import os import numpy as np import scipy.misc import tensorflow as tf from tensorflow.examples.tutorials.mnist import input_data #as mnist_data def conv2d(name, tensor,ksize, out_dim, stddev=0.01, stride=2, padding='SAME'): with tf.variable_scope(name): w = tf.get_variable('w', [ksize, ksize, tensor.get_shape()[-1],out_dim], dtype=tf.float32, initializer=tf.random_normal_initializer(stddev=stddev)) var = tf.nn.conv2d(tensor,w,[1,stride, stride,1],padding=padding) b = tf.get_variable('b', [out_dim], 'float32',initializer=tf.constant_initializer(0.01)) return tf.nn.bias_add(var, b) def deconv2d(name, tensor, ksize, outshape, stddev=0.01, stride=2, padding='SAME'): with tf.variable_scope(name): w = tf.get_variable('w', [ksize, ksize, outshape[-1], tensor.get_shape()[-1]], dtype=tf.float32, initializer=tf.random_normal_initializer(stddev=stddev)) var = tf.nn.conv2d_transpose(tensor, w, outshape, strides=[1, stride, stride, 1], padding=padding) b = tf.get_variable('b', [outshape[-1]], 'float32', initializer=tf.constant_initializer(0.01)) return tf.nn.bias_add(var, b) def fully_connected(name,value, output_shape): with tf.variable_scope(name, reuse=None) as scope: shape = value.get_shape().as_list() w = tf.get_variable('w', [shape[1], output_shape], dtype=tf.float32, initializer=tf.random_normal_initializer(stddev=0.01)) b = tf.get_variable('b', [output_shape], dtype=tf.float32, initializer=tf.constant_initializer(0.0)) return tf.matmul(value, w) + b def relu(name, tensor): return tf.nn.relu(tensor, name) def lrelu(name,x, leak=0.2): return tf.maximum(x, leak * x, name=name) DEPTH = 28 OUTPUT_SIZE = 28 batch_size = 64 def Discriminator(name,inputs,reuse): with tf.variable_scope(name, reuse=reuse): output = tf.reshape(inputs, [-1, 28, 28, 1]) output1 = conv2d('d_conv_1', output, ksize=5, out_dim=DEPTH) output2 = lrelu('d_lrelu_1', output1) output3 = conv2d('d_conv_2', output2, ksize=5, out_dim=2*DEPTH) output4 = lrelu('d_lrelu_2', output3) output5 = conv2d('d_conv_3', output4, ksize=5, out_dim=4*DEPTH) output6 = lrelu('d_lrelu_3', output5) # output7 = conv2d('d_conv_4', output6, ksize=5, out_dim=8*DEPTH) # output8 = lrelu('d_lrelu_4', output7) chanel = output6.get_shape().as_list() output9 = tf.reshape(output6, [batch_size, chanel[1]*chanel[2]*chanel[3]]) output0 = fully_connected('d_fc', output9, 1) return output0 def generator(name, reuse=False): with tf.variable_scope(name, reuse=reuse): noise = tf.random_normal([batch_size, 128])#.astype('float32') noise = tf.reshape(noise, [batch_size, 128], 'noise') output = fully_connected('g_fc_1', noise, 2*2*8*DEPTH) output = tf.reshape(output, [batch_size, 2, 2, 8*DEPTH], 'g_conv') output = deconv2d('g_deconv_1', output, ksize=5, outshape=[batch_size, 4, 4, 4*DEPTH]) output = tf.nn.relu(output) output = tf.reshape(output, [batch_size, 4, 4, 4*DEPTH]) output = deconv2d('g_deconv_2', output, ksize=5, outshape=[batch_size, 7, 7, 2* DEPTH]) output = tf.nn.relu(output) output = deconv2d('g_deconv_3', output, ksize=5, outshape=[batch_size, 14, 14, DEPTH]) output = tf.nn.relu(output) output = deconv2d('g_deconv_4', output, ksize=5, outshape=[batch_size, OUTPUT_SIZE, OUTPUT_SIZE, 1]) # output = tf.nn.relu(output) output = tf.nn.sigmoid(output) return tf.reshape(output,[-1,784]) def save_images(images, size, path): # 图片归一化 img = (images + 1.0) / 2.0 h, w = img.shape[1], img.shape[2] merge_img = np.zeros((h * size[0], w * size[1], 3)) for idx, image in enumerate(images): i = idx % size[1] j = idx // size[1] merge_img[j * h:j * h + h, i * w:i * w + w, :] = image return scipy.misc.imsave(path, merge_img) LAMBDA = 10 EPOCH = 40 def train(): # print os.getcwd() with tf.variable_scope(tf.get_variable_scope()): # real_data = tf.placeholder(dtype=tf.float32, shape=[-1, OUTPUT_SIZE*OUTPUT_SIZE*3]) path = os.getcwd() data_dir = path + "/train.tfrecords"#准备使用自己的数据集 # print data_dir '''获得数据''' z = tf.placeholder(dtype=tf.float32, shape=[batch_size, 100])#build placeholder real_data = tf.placeholder(tf.float32, shape=[batch_size,784]) with tf.variable_scope(tf.get_variable_scope()): fake_data = generator('gen',reuse=False) disc_real = Discriminator('dis_r',real_data,reuse=False) disc_fake = Discriminator('dis_r',fake_data,reuse=True) t_vars = tf.trainable_variables() d_vars = [var for var in t_vars if 'd_' in var.name] g_vars = [var for var in t_vars if 'g_' in var.name] '''计算损失''' gen_cost = tf.reduce_mean(disc_fake) disc_cost = -tf.reduce_mean(disc_fake) + tf.reduce_mean(disc_real) alpha = tf.random_uniform( shape=[batch_size, 1],minval=0.,maxval=1.) differences = fake_data - real_data interpolates = real_data + (alpha * differences) gradients = tf.gradients(Discriminator('dis_r',interpolates,reuse=True), [interpolates])[0] slopes = tf.sqrt(tf.reduce_sum(tf.square(gradients), reduction_indices=[1])) gradient_penalty = tf.reduce_mean((slopes - 1.) ** 2) disc_cost += LAMBDA * gradient_penalty with tf.variable_scope(tf.get_variable_scope(), reuse=None): gen_train_op = tf.train.AdamOptimizer( learning_rate=1e-4,beta1=0.5,beta2=0.9).minimize(gen_cost,var_list=g_vars) disc_train_op = tf.train.AdamOptimizer( learning_rate=1e-4,beta1=0.5,beta2=0.9).minimize(disc_cost,var_list=d_vars) saver = tf.train.Saver() # os.environ['CUDA_VISIBLE_DEVICES'] = str(0)#gpu环境 # config = tf.ConfigProto() # config.gpu_options.per_process_gpu_memory_fraction = 0.5#调用50%GPU资源 # sess = tf.InteractiveSession(config=config) sess = tf.InteractiveSession() coord = tf.train.Coordinator() threads = tf.train.start_queue_runners(sess=sess, coord=coord) if not os.path.exists('img'): os.mkdir('img') init = tf.global_variables_initializer() # init = tf.initialize_all_variables() sess.run(init) mnist = input_data.read_data_sets("MNIST_data", one_hot=True) # mnist = mnist_data.read_data_sets("data", one_hot=True, reshape=False, validation_size=0) for epoch in range (1, EPOCH): idxs = 1000 for iters in range(1, idxs): img, _ = mnist.train.next_batch(batch_size) # img2 = tf.reshape(img, [batch_size, 784]) for x in range (0,5): _, d_loss = sess.run([disc_train_op, disc_cost], feed_dict={real_data: img}) _, g_loss = sess.run([gen_train_op, gen_cost]) # print "fake_data:%5f disc_real:%5f disc_fake:%5f "%(tf.reduce_mean(fake_data) # ,tf.reduce_mean(disc_real),tf.reduce_mean(disc_fake)) print("[%4d:%4d/%4d] d_loss: %.8f, g_loss: %.8f"%(epoch, iters, idxs, d_loss, g_loss)) with tf.variable_scope(tf.get_variable_scope()): samples = generator('gen', reuse=True) samples = tf.reshape(samples, shape=[batch_size, 28,28,1]) samples=sess.run(samples) save_images(samples, [8,8], os.getcwd()+'/img/'+'sample_%d_epoch.png' % (epoch)) if epoch>=39: checkpoint_path = os.path.join(os.getcwd(), 'my_wgan-gp.ckpt') saver.save(sess, checkpoint_path, global_step=epoch) print ('********* model saved *********') coord.request_stop() coord.join(threads) sess.close() if __name__ == '__main__': train()
import tensorflow as tf from tensorflow.examples.tutorials.mnist import input_data import os import numpy as np from scipy import misc,ndimage import matplotlib.pyplot as plt from bokeh.charts.attributes import color mnist = input_data.read_data_sets('./MNIST_data') batch_size = 100 width,height = 28,28 mnist_dim = width*height random_dim = 10 epochs = 1000000 def my_init(size): return tf.random_uniform(size, -0.05, 0.05) D_W1 = tf.Variable(my_init([mnist_dim, 128])) D_b1 = tf.Variable(tf.zeros([128])) D_W2 = tf.Variable(my_init([128, 32])) D_b2 = tf.Variable(tf.zeros([32])) D_W3 = tf.Variable(my_init([32, 1])) D_b3 = tf.Variable(tf.zeros([1])) D_variables = [D_W1, D_b1, D_W2, D_b2, D_W3, D_b3] G_W1 = tf.Variable(my_init([random_dim, 32])) G_b1 = tf.Variable(tf.zeros([32])) G_W2 = tf.Variable(my_init([32, 128])) G_b2 = tf.Variable(tf.zeros([128])) G_W3 = tf.Variable(my_init([128, mnist_dim])) G_b3 = tf.Variable(tf.zeros([mnist_dim])) G_variables = [G_W1, G_b1, G_W2, G_b2, G_W3, G_b3] def D(X): X = tf.nn.relu(tf.matmul(X, D_W1) + D_b1) X = tf.nn.relu(tf.matmul(X, D_W2) + D_b2) X = tf.matmul(X, D_W3) + D_b3 return X def G(X): X = tf.nn.relu(tf.matmul(X, G_W1) + G_b1) X = tf.nn.relu(tf.matmul(X, G_W2) + G_b2) X = tf.nn.sigmoid(tf.matmul(X, G_W3) + G_b3) return X real_X = tf.placeholder(tf.float32, shape=[batch_size, mnist_dim]) random_X = tf.placeholder(tf.float32, shape=[batch_size, random_dim]) random_Y = G(random_X) eps = tf.random_uniform([batch_size, 1], minval=0., maxval=1.) X_inter = eps*real_X + (1. - eps)*random_Y grad = tf.gradients(D(X_inter), [X_inter])[0] grad_norm = tf.sqrt(tf.reduce_sum((grad)**2, axis=1)) grad_pen = 10 * tf.reduce_mean(tf.nn.relu(grad_norm - 1.)) D_loss = tf.reduce_mean(D(real_X)) - tf.reduce_mean(D(random_Y)) + grad_pen G_loss = tf.reduce_mean(D(random_Y)) D_solver = tf.train.AdamOptimizer(1e-4, 0.5).minimize(D_loss, var_list=D_variables) G_solver = tf.train.AdamOptimizer(1e-4, 0.5).minimize(G_loss, var_list=G_variables) sess = tf.Session() sess.run(tf.global_variables_initializer()) if not os.path.exists('out/'): os.makedirs('out/') for e in range(epochs): for i in range(5): real_batch_X,_ = mnist.train.next_batch(batch_size) random_batch_X = np.random.uniform(-1, 1, (batch_size, random_dim)) _,D_loss_ = sess.run([D_solver,D_loss], feed_dict={real_X:real_batch_X, random_X:random_batch_X}) random_batch_X = np.random.uniform(-1, 1, (batch_size, random_dim)) _,G_loss_ = sess.run([G_solver,G_loss], feed_dict={random_X:random_batch_X}) if e % 10 == 0: print ('epoch %s, D_loss: %s, G_loss: %s'%(e, D_loss_, G_loss_)) n_rows = 6 if e % 1000 == 0: check_imgs = sess.run(random_Y, feed_dict={random_X:random_batch_X}).reshape((100, width, height)) r,c = 10,10 cnt = 0 fig,axs = plt.subplots(r,c) for i in range(r): for j in range(c): axs[i,j].imshow(check_imgs[cnt,:,:],cmap='gray') axs[i,j].axis('off') cnt+=1 plt.show() # imgs = np.ones((width*n_rows+5*n_rows+5, height*n_rows+5*n_rows+5)) # for i in range(n_rows*n_rows): # imgs[5+5*(i%n_rows)+width*(i%n_rows):5+5*(i%n_rows)+width+width*(i%n_rows), 5+5*(i/n_rows)+height*(i/n_rows):5+5*(i/n_rows)+height+height*(i/n_rows)] = check_imgs[i] # misc.imsave('out/%s.png'%(e/1000), imgs)
WGAN-GP生成自己的数据
#coding:utf-8 import os import numpy as np import scipy.misc import tensorflow as tf from six.moves import xrange import matplotlib.pyplot as plt def conv2d(name, tensor,ksize, out_dim, stddev=0.01, stride1=1,stride2=1, padding='SAME'): with tf.variable_scope(name): w = tf.get_variable('w', [ksize, ksize, tensor.get_shape()[-1],out_dim], dtype=tf.float32, initializer=tf.random_normal_initializer(stddev=stddev)) var = tf.nn.conv2d(tensor,w,[1,stride1, stride2,1],padding=padding) b = tf.get_variable('b', [out_dim], 'float32',initializer=tf.constant_initializer(0.01)) return tf.nn.bias_add(var, b) def deconv2d(name, tensor, ksize, outshape, stddev=0.01, stride1= 1,stride2=1, padding='SAME'): with tf.variable_scope(name): w = tf.get_variable('w', [ksize, ksize, outshape[-1], tensor.get_shape()[-1]], dtype=tf.float32, initializer=tf.random_normal_initializer(stddev=stddev)) var = tf.nn.conv2d_transpose(tensor, w, outshape, strides=[1, stride1, stride2, 1], padding=padding) b = tf.get_variable('b', [outshape[-1]], 'float32', initializer=tf.constant_initializer(0.01)) return tf.nn.bias_add(var, b) def fully_connected(name,value, output_shape): with tf.variable_scope(name, reuse=None) as scope: shape = value.get_shape().as_list() w = tf.get_variable('w', [shape[1], output_shape], dtype=tf.float32, initializer=tf.random_normal_initializer(stddev=0.01)) b = tf.get_variable('b', [output_shape], dtype=tf.float32, initializer=tf.constant_initializer(0.0)) return tf.matmul(value, w) + b def relu(name, tensor): return tf.nn.relu(tensor, name) def lrelu(name,x, leak=0.2): return tf.maximum(x, leak * x, name=name) width = 3 height = 60 batch_size = 100 a =32 b=5 def Discriminator(name,inputs,reuse): with tf.variable_scope(name, reuse=reuse): output = tf.reshape(inputs, [-1, 3, 60, 1]) print(output.shape) output1 = conv2d('d_conv_1', output, ksize=b, out_dim=a) output2 = lrelu('d_lrelu_1', output1) print(output2.shape) output3 = conv2d('d_conv_2', output2, ksize=b, out_dim=2*a) output4 = lrelu('d_lrelu_2', output3) print(output4.shape) output5 = conv2d('d_conv_3', output4, ksize=b, out_dim=4*a) output6 = lrelu('d_lrelu_3', output5) print(output6.shape) # output7 = conv2d('d_conv_4', output6, ksize=5, out_dim=8*width) # output8 = lrelu('d_lrelu_4', output7) chanel = output6.get_shape().as_list() output9 = tf.reshape(output6, [batch_size, chanel[1]*chanel[2]*chanel[3]]) print(output9.shape) output0 = fully_connected('d_fc', output9, 1) return output0 def generator(name, reuse=False): with tf.variable_scope(name, reuse=reuse): noise = tf.random_normal([batch_size, 100])#.astype('float32') noise = tf.reshape(noise, [batch_size, 100], 'noise') output = fully_connected('g_fc_1', noise, 3*60*8*a) output = tf.reshape(output, [batch_size, 3, 60, 8*a], 'g_conv') print(output.shape) output = deconv2d('g_deconv_1', output, ksize=b, outshape=[batch_size, 3, 60, 4*a]) output = tf.nn.relu(output) output = tf.reshape(output, [batch_size, 3, 60, 4*a]) print(output.shape) output = deconv2d('g_deconv_2', output, ksize=b, outshape=[batch_size, 3, 60, 2* a]) output = tf.nn.relu(output) output = deconv2d('g_deconv_3', output, ksize=b, outshape=[batch_size, 3, 60, a]) output = tf.nn.relu(output) output = deconv2d('g_deconv_4', output, ksize=b, outshape=[batch_size, 3, 60, 1]) print(output.shape) # output = tf.nn.relu(output) output = tf.nn.sigmoid(output) return tf.reshape(output,[-1,180]) def save_images(images, size, path): # 图片归一化 img = (images + 1.0) / 2.0 h, w = img.shape[1], img.shape[2] merge_img = np.zeros((h * size[0], w * size[1], 3)) for idx, image in enumerate(images): i = idx % size[1] j = idx // size[1] merge_img[j * h:j * h + h, i * w:i * w + w, :] = image return scipy.misc.imsave(path, merge_img) LAMBDA = 10 EPOCH = 40 def train(): # print os.getcwd() with tf.variable_scope(tf.get_variable_scope()): # real_data = tf.placeholder(dtype=tf.float32, shape=[-1, height*height*3]) path = os.getcwd() data_dir = path + "/train.tfrecords"#准备使用自己的数据集 # print data_dir '''获得数据''' z = tf.placeholder(dtype=tf.float32, shape=[batch_size, 100])#build placeholder real_data = tf.placeholder(tf.float32, shape=[batch_size,180]) with tf.variable_scope(tf.get_variable_scope()): fake_data = generator('gen',reuse=False) disc_real = Discriminator('dis_r',real_data,reuse=False) disc_fake = Discriminator('dis_r',fake_data,reuse=True) t_vars = tf.trainable_variables() d_vars = [var for var in t_vars if 'd_' in var.name] g_vars = [var for var in t_vars if 'g_' in var.name] '''计算损失''' gen_cost = tf.reduce_mean(disc_fake) disc_cost = -tf.reduce_mean(disc_fake) + tf.reduce_mean(disc_real) alpha = tf.random_uniform( shape=[batch_size, 1],minval=0.,maxval=1.) differences = fake_data - real_data interpolates = real_data + (alpha * differences) gradients = tf.gradients(Discriminator('dis_r',interpolates,reuse=True), [interpolates])[0] slopes = tf.sqrt(tf.reduce_sum(tf.square(gradients), reduction_indices=[1])) gradient_penalty = tf.reduce_mean((slopes - 1.) ** 2) disc_cost += LAMBDA * gradient_penalty with tf.variable_scope(tf.get_variable_scope(), reuse=None): gen_train_op = tf.train.AdamOptimizer( learning_rate=1e-4,beta1=0.5,beta2=0.9).minimize(gen_cost,var_list=g_vars) disc_train_op = tf.train.AdamOptimizer( learning_rate=1e-4,beta1=0.5,beta2=0.9).minimize(disc_cost,var_list=d_vars) saver = tf.train.Saver() # os.environ['CUDA_VISIBLE_DEVICES'] = str(0)#gpu环境 # config = tf.ConfigProto() # config.gpu_options.per_process_gpu_memory_fraction = 0.5#调用50%GPU资源 # sess = tf.InteractiveSession(config=config) sess = tf.InteractiveSession() coord = tf.train.Coordinator() threads = tf.train.start_queue_runners(sess=sess, coord=coord) if not os.path.exists('img'): os.mkdir('img') init = tf.global_variables_initializer() # init = tf.initialize_all_variables() sess.run(init) data = np.load('data/final37.npy') # print(data.shape) data = data[:,:,0:60] for epoch in range (1, EPOCH): idxs = 1000 for iters in range(1, idxs): X_train = data idx = np.random.randint(0,X_train.shape[0],batch_size) img = X_train[idx] img = img.reshape(-1,180) for x in range (0,5): _, d_loss = sess.run([disc_train_op, disc_cost], feed_dict={real_data: img}) _, g_loss = sess.run([gen_train_op, gen_cost]) print("[%4d:%4d/%4d] d_loss: %.8f, g_loss: %.8f"%(epoch, iters, idxs, d_loss, g_loss)) with tf.variable_scope(tf.get_variable_scope()): samples = generator('gen', reuse=True) samples = tf.reshape(samples, shape=[batch_size, 3,60,1]) samples=sess.run(samples) gen = samples r, c = 10, 10 fig, axs = plt.subplots(r, c) cnt = 0 for i in range(r): for j in range(c): xy = gen[cnt]#第n个分叉图,有三个分支,每个分支21个数 for k in range(len(xy)): x = xy[k][0:30] y = xy[k][30:60] axs[i,j].plot(x,y) axs[i,j].axis('off') cnt += 1 if not os.path.exists('wgan-gp'): os.makedirs('wgan-gp') fig.savefig("wgan-gp/%d.png" % epoch) plt.close() if epoch>=39: checkpoint_path = os.path.join(os.getcwd(), 'my_wgan-gp.ckpt') saver.save(sess, checkpoint_path, global_step=epoch) print ('********* model saved *********') coord.request_stop() coord.join(threads) sess.close() if __name__ == '__main__': train()