def model_nn(sess,
input_image,
num_iterations=200,
is_print_info=True,
is_plot=True,
is_save_process_image=True,
save_last_image_to="output/generated_image.jpg"):
#初始化全局变量
sess.run(tf.global_variables_initializer())
#运行带噪声的输入图像
sess.run(model["input"].assign(input_image))
for i in range(num_iterations):
#运行最小化的目标：
sess.run(train_step)
#产生把数据输入模型后生成的图像
generated_image = sess.run(model["input"])
if is_print_info and i % 20 == 0:
Jt, Jc, Js = sess.run([J, J_content, J_style])
print("第 " + str(i) + "轮训练," + " 总成本为:" + str(Jt) + " 内容成本为：" +
str(Jc) + " 风格成本为：" + str(Js))
if is_save_process_image:
nst_utils.save_image("output/" + str(i) + ".png", generated_image)
nst_utils.save_image(save_last_image_to, generated_image)
return generated_image
def NST_model(num_iter=1000):
content_img = load_img("con_niu.jpg")
content_img = nu.reshape_and_normalize_image(content_img)
style_img = load_img("style_cloud.jpg")
style_img = nu.reshape_and_normalize_image(style_img)
generated_img = nu.generate_noise_image(content_img)
print(np.shape(content_img))
print(np.shape(style_img))
print(np.shape(generated_img))
with tf.Session() as sess:
model = nu.load_vgg_model(
"pretrained-model/imagenet-vgg-verydeep-19.mat")
sess.run(model['input'].assign(content_img))
out = model['conv4_2']
a_c = sess.run(out)
a_g = out
J_content = content_cost(a_c, a_g)
STYLE_LAYERS = [('conv1_1', 0.2), ('conv2_1', 0.2), ('conv3_1', 0.2),
('conv4_1', 0.2), ('conv5_1', 0.2)]
sess.run(model['input'].assign(style_img))
J_style = style_cost(model, STYLE_LAYERS, sess)
J = total_cost(J_content, J_style)
optimizer = tf.train.AdamOptimizer(2.0)
train_step = optimizer.minimize(J)
tf.global_variables_initializer().run()
sess.run(model['input'].assign(generated_img))
for i in range(num_iter):
sess.run(train_step)
generated_img = sess.run(model['input'])
if i % 20 == 0:
Jt, Jc, Js = sess.run([J, J_content, J_style])
print("Iteration " + str(i) + " :")
print("total cost = " + str(Jt))
print("content cost = " + str(Jc))
print("style cost = " + str(Js))
print(generated_img.shape)
nu.save_image("output4/" + str(i) + ".png", generated_img)
nu.save_image("output4/generated_image.png", generated_img)
return generated_img
def nn_model(sess, input_image, num_iterations=300):
sess.run(tf.global_variables_initializer())
sess.run(model['input'].assign(input_image))
for i in range(num_iterations):
sess.run(train_step)
generated_image = sess.run(model['input'])
if i % 20 == 0:
Jt, Jc, Js = sess.run([J, J_content, J_style])
print("Iteration " + str(i) + " :")
print("total cost = " + str(Jt))
print("content cost = " + str(Jc))
print("style cost = " + str(Js))
save_image("./a_" + str(i) + ".png", generated_image)
save_image('./generated_image_2.jpg', generated_image)
return generated_image
def model_nst(self,
input_image,
optimizer,
content_img,
style_img,
output_name,
graph,
sess,
print_output=False):
sess.run(self.model['input'].assign(content_img))
out = self.model['conv4_2']
a_c = sess.run(out)
a_g = out
j_content = self.compute_content_cost(a_c, a_g)
sess.run(self.model['input'].assign(style_img))
j_style = self.compute_style_cost(self.model, self.style_layers, sess)
j = self.total_cost(j_content, j_style)
train_step = optimizer.minimize(j)
sess.run(tf.global_variables_initializer())
sess.run(self.model['input'].assign(input_image))
for i in range(self.num_iterations):
sess.run(train_step)
generated_image = sess.run(self.model['input'])
if i % 20 == 0 and print_output:
jt, jc, js = sess.run([j, j_content, j_style])
print("Iteration " + str(i) + " :")
print("Total cost = " + str(jt))
print("Content cost = " + str(jc))
print("Style cost = " + str(js))
save_image(
self.output_dir + output_name + '_' + str(i) + ".png",
generated_image)
sess.close()
path_generated_img = self.output_dir + output_name + '.png'
save_image(path_generated_img, generated_image)
return generated_image, path_generated_img
def fit(self, learning_rate=2.0, num_iterations=200):
history = []
# Assign the content image to be the input of the VGG model.
# sess.run(self.model['input'].assign(self.content_image))
# out = self.model['conv4_2']
# a_C = sess.run(out)
# a_G = out
# J_content = self.compute_content_cost(a_C, a_G)
J_content = self.content_cost()
J_style = self.style_cost()
J = self.total_cost(J_content, J_style)
optimizer = tf.train.AdamOptimizer(learning_rate)
self.train_step = optimizer.minimize(J)
# init global variables
sess.run(tf.global_variables_initializer())
# run the noisy input image, use assign()
sess.run(self.model['input'].assign(self.init_image))
if not os.path.isdir('output'):
os.mkdir('output')
for i in range(num_iterations):
sess.run(self.train_step)
generated_image = sess.run(self.model['input'])
# Print every 20 iteration.
if i % 20 == 0:
Jt, Jc, Js = sess.run([J, J_content, J_style])
print("Iteration " + str(i) + " :")
print("total cost = " + str(Jt))
print("content cost = " + str(Jc))
print("style cost = " + str(Js))
# save current generated image in the "/output" directory
nst_utils.save_image("output/" + str(i) + ".png",
generated_image)
history.append((Jt, Jc, Js))
# save last generated image
nst_utils.save_image('output/generated_image.jpg', generated_image)
return history
def model_nn(sess, input_image, num_iterations=400):
# Initialize global variables (you need to run the session on the initializer)
### START CODE HERE ### (1 line)
sess.run(tf.global_variables_initializer())
### END CODE HERE ###
# Run the noisy input image (initial generated image) through the model. Use assign().
### START CODE HERE ### (1 line)
generated_image = input_image
sess.run(model['input'].assign(input_image))
### END CODE HERE ###
for i in range(num_iterations):
# Run the session on the train_step to minimize the total cost
### START CODE HERE ### (1 line)
sess.run(train_step)
### END CODE HERE ###
# Compute the generated image by running the session on the current model['input']
### START CODE HERE ### (1 line)
generated_image = sess.run(model['input'])
### END CODE HERE ###
# Print every 20 iteration.
if i % 20 == 0:
Jt, Jc, Js = sess.run([J, J_content, J_style])
print("Iteration " + str(i) + " :")
print("total cost = " + str(Jt))
print("content cost = " + str(Jc))
print("style cost = " + str(Js))
# save current generated image in the "/output" directory
save_image("output/" + str(i) + ".png", generated_image)
# save last generated image
save_image('output/generated_image.jpg', generated_image)
return generated_image
def generate_noise_image2(content_image, noise_ratio=0.6):
"""
Generates a noisy image by adding random noise to the content_image
"""
height = content_image.shape[1]
width = content_image.shape[2]
# Generate a random noise_image
noise_image = np.random.uniform(-20, 20,
(1, height, width, 3)).astype('float32')
# range_ = np.max(content_image) - np.min(content_image)
# noise_image = np.random.normal(loc=np.mean(content_image), scale = 0.7*range_, size=(1, height, width, 3)).astype('float32')
# Set the input_image to be a weighted average of the content_image and a noise_image
input_image = noise_image * noise_ratio + content_image * (1 - noise_ratio)
return input_image
STYLE_LAYERS = [('conv1_1', 0.2), ('conv2_1', 0.2), ('conv3_1', 0.2),
('conv4_1', 0.2), ('conv5_1', 0.2)]
# Reset the graph
tf.reset_default_graph()
# Start interactive session
sess = tf.InteractiveSession()
content_image = scipy.misc.imread("images/z_me3.jpg")
content_image = reshape_and_normalize_image(content_image)
style_image = scipy.misc.imread("images/constable.jpg")
style_image = reshape_and_normalize_image(style_image)
generated_image = generate_noise_image2(content_image)
#print(generated_image.shape)
imshow(generated_image[0])
model = load_vgg_model("pretrained-model/imagenet-vgg-verydeep-19.mat")
# Assign the content image to be the input of the VGG model.
sess.run(model['input'].assign(content_image))
# Select the output tensor of layer conv4_2
out = model['conv4_2']
# Set a_C to be the hidden layer activation from the layer we have selected
a_C = sess.run(out)
# Set a_G to be the hidden layer activation from same layer. Here, a_G references model['conv4_2']
# and isn't evaluated yet. Later in the code, we'll assign the image G as the model input, so that
# when we run the session, this will be the activations drawn from the appropriate layer, with G as input.
a_G = out
# Compute the content cost
J_content = compute_content_cost(a_C, a_G)
# Assign the input of the model to be the "style" image
sess.run(model['input'].assign(style_image))
# Compute the style cost
J_style = compute_style_cost(sess, model, STYLE_LAYERS)
### START CODE HERE ### (1 line)
J = total_cost(J_content, J_style, alpha=10, beta=40)
### END CODE HERE ###
# define optimizer (1 line)
optimizer = tf.train.AdamOptimizer(1.0)
# define train_step (1 line)
train_step = optimizer.minimize(J)
# Initialize global variables (you need to run the session on the initializer)
### START CODE HERE ### (1 line)
sess.run(tf.global_variables_initializer())
### END CODE HERE ###
# Run the noisy input image (initial generated image) through the model. Use assign().
### START CODE HERE ### (1 line)
sess.run(model['input'].assign(generated_image))
### END CODE HERE ###
for i in range(200):
# Run the session on the train_step to minimize the total cost
### START CODE HERE ### (1 line)
sess.run(train_step)
### END CODE HERE ###
# Compute the generated image by running the session on the current model['input']
### START CODE HERE ### (1 line)
generated_image = sess.run(model['input'])
### END CODE HERE ###
# Print every 20 iteration.
if i % 2 == 0:
Jt, Jc, Js = sess.run([J, J_content, J_style])
print("Iteration " + str(i) + " :")
print("total cost = " + str(Jt))
print("content cost = " + str(Jc))
print("style cost = " + str(Js))
# save current generated image in the "/output" directory
save_image("output/z_m3_3_constable_" + str(i) + ".png",
generated_image)
# save last generated image
save_image('output/z_me_3_constable_generated_image.jpg', generated_image)
return generated_image
def Generate(cls,
content,
style,
alpha=10,
beta=40,
no_iter=100,
display=False):
"""
call signature : Generate(content,style,alpha=10,beta=40,iter=100)
input -- content : content image,
style : style image,
alpha : content cost multiplier,
beta : style cost multiplier,
iter : number of iteration
return --- a dictionary
total_cost : array of total cost at eact 10th iteration,
content_cost : array of content cost at each 10th iteration,
style_cost : array of style cost at each 10th iteration,
image : np array of generated image of shape(1,h,w,c)
side effect ---
save the generated image in output dir
"""
J_content = cls.calculateTotalContentCost(content, cls.content)
J_style = cls.calculateTotalStyleCost(style, cls.style)
#compute Total Cost
J = alpha * J_content + beta * J_style
#Initialize noisy Generated Image
image = generate_noise_image(content)
#Set Optimizer
optimizer = tf.train.AdamOptimizer(cls.learning_rate)
train_step = optimizer.minimize(J)
#initialize plot
if (display):
fig, ax1, ax2 = cls._createFig()
J_show = []
J_C_show = []
J_S_show = []
#Start the Session
with tf.Session() as sess:
#initialize Variables
sess.run(tf.global_variables_initializer())
sess.run(model['input'].assign(image))
#run optimization
for i in range(no_iter):
sess.run(train_step)
generated_image = sess.run(model['input'])
#print infomation
if i % 10 == 0:
temp_1, temp_2, temp_3 = sess.run([J, J_content, J_style])
J_show.append(temp_1)
J_C_show.append(temp_2)
J_S_show.append(temp_3)
print('iter : {}, J : {}'.format(i, temp_1))
if (display):
cls._updateFig(fig, ax1, ax2, generated_image,
(J_show, J_C_show, J_S_show), no_iter)
mat = {
'total_cost': J_show,
'content_cost': J_C_show,
'style_cost': J_S_show,
'image': generated_image
}
save_image(CONFIG.OUTPUT_DIR + cls.name + '.jpg', generated_image)
scipy.io.savemat(CONFIG.OUTPUT_DIR + cls.name + '.mat', mat)
return mat
optimizer = tf.train.AdamOptimizer(CONFIG.LEARNING_RATE)
train_step = optimizer.minimize(J)
# Step 7: Run graph for a large number of iterations, updating the generated image at every step
# Initialize global variable
sess.run(tf.global_variables_initializer())
# Run the noisy initial generated image through the model.
sess.run(model['input'].assign(generated_image))
for i in range(CONFIG.NUM_ITERATIONS):
# Run the session on the train_step to minimize the total cost
sess.run(train_step)
# Compute the generated image by running the session on the current model['input']
generated_image = sess.run(model['input'])
# Print every 20 iteration.
if i % 20 == 0:
Jt, Jc, Js = sess.run([J, J_content, J_style])
print("Iteration " + str(i) + " :")
print("total cost = " + str(Jt))
print("content cost = " + str(Jc))
print("style cost = " + str(Js))
# save current generated image in the "/output" directory
save_image("output/" + str(i) + ".png", generated_image)
# save last generated image
save_image('output/generated_image.jpg', generated_image)
