cgan_fashionmnist_tensorflow.py

import imageio
import glob
import os
import time
import cv2
import tensorflow as tf
from tensorflow.keras import layers
from IPython import display
import matplotlib.pyplot as plt
import numpy as np
from tensorflow.keras import backend as K
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt
from tensorflow import keras
from matplotlib import pyplot
from numpy import asarray
from numpy.random import randn
from numpy.random import randint
from numpy import linspace
from matplotlib import pyplot

(x_train, y_train), (x_test, y_test) = tf.keras.datasets.fashion_mnist.load_data()
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1).astype('float32')
x_test = x_test.astype('float32')
x_train = (x_train / 127.5) - 1 
# Batch and shuffle the data
train_dataset = tf.data.Dataset.from_tensor_slices((x_train,y_train)).\
shuffle(60000).batch(128)

plt.figure(figsize=(10, 10))
for images,_ in train_dataset.take(1):
    for i in range(100):
        ax = plt.subplot(10, 10, i + 1)
        plt.imshow(images[i,:,:,0].numpy().astype("uint8"), cmap='gray')
        plt.axis("off")

BATCH_SIZE=128
latent_dim = 100

# label input
con_label = layers.Input(shape=(1,))

# image generator input
latent_vector = layers.Input(shape=(100,))

def label_conditioned_gen(n_classes=10, embedding_dim=100):
    # embedding for categorical input
    label_embedding = layers.Embedding(n_classes, embedding_dim)(con_label)
    # linear multiplication
    n_nodes = 7 * 7
    label_dense = layers.Dense(n_nodes)(label_embedding)
    # reshape to additional channel
    label_reshape_layer = layers.Reshape((7, 7, 1))(label_dense)
    return label_reshape_layer

def latent_gen(latent_dim=100):
    # image generator input
    in_lat = layers.Input(shape=(latent_dim,))
    n_nodes = 128 * 7 * 7
    latent_dense = layers.Dense(n_nodes)(latent_vector)
    latent_dense = layers.LeakyReLU(alpha=0.2)(latent_dense)
    latent_reshape = layers.Reshape((7, 7, 128))(latent_dense)
    return latent_reshape

def con_generator():
    latent_vector_output = label_conditioned_gen()
    label_output = latent_gen()
    # merge image gen and label input
    merge = layers.Concatenate()([latent_vector_output, label_output])
    # upsample to 14x14
    x = layers.Conv2DTranspose(128, (4,4), strides=(2,2), padding='same')(merge)
    x = layers.LeakyReLU(alpha=0.2)(x)
    # upsample to 28x28
    x = layers.Conv2DTranspose(128, (4,4), strides=(2,2), padding='same')(x)
    x = layers.LeakyReLU(alpha=0.2)(x)
    # output
    out_layer = layers.Conv2D(1, (7,7), activation='tanh', padding='same')(x)
    # define model
    model = tf.keras.Model([latent_vector, con_label], out_layer)
    return model

conditional_gen = con_generator()

conditional_gen.summary()

def label_condition_disc(in_shape=(28, 28, 1), n_classes=10, embedding_dim=100):
    # label input
    con_label = layers.Input(shape=(1,))
    # embedding for categorical input
    label_embedding = layers.Embedding(n_classes, embedding_dim)(con_label)
    # scale up to image dimensions with linear activation
    nodes = in_shape[0] * in_shape[1] * in_shape[2]
    label_dense = layers.Dense(nodes)(label_embedding)
    # reshape to additional channel
    label_reshape_layer = layers.Reshape((in_shape[0], in_shape[1], 1))(label_dense)
    # image input
    return con_label, label_reshape_layer


def image_disc(in_shape=(28,28, 1)):
    inp_image = layers.Input(shape=in_shape)
    return inp_image 

def con_discriminator():
    con_label, label_condition_output = label_condition_disc()
    inp_image_output = image_disc()
    # concat label as a channel
    merge = layers.Concatenate()([inp_image_output, label_condition_output])
    # downsample
    x = layers.Conv2D(128, (3,3), strides=(2,2), padding='same')(merge)
    x = layers.LeakyReLU(alpha=0.2)(x)
    # downsample
    x = layers.Conv2D(128, (3,3), strides=(2,2), padding='same')(x)
    x = layers.LeakyReLU(alpha=0.2)(x)
    # flatten feature maps
    flattened_out = layers.Flatten()(x)
    # dropout
    dropout = layers.Dropout(0.4)(flattened_out)
    # output
    dense_out = layers.Dense(1, activation='sigmoid')(dropout)
    # define model
    model = tf.keras.Model([inp_image_output, con_label], dense_out)
    return model

conditional_discriminator = con_discriminator()

conditional_discriminator.summary()

binary_cross_entropy = tf.keras.losses.BinaryCrossentropy()

def generator_loss(label, fake_output):
    gen_loss = binary_cross_entropy(label, fake_output)
    #print(gen_loss)
    return gen_loss

def discriminator_loss(label, output):
    disc_loss = binary_cross_entropy(label, output)
    #print(total_loss)
    return disc_loss

learning_rate = 0.0002 
generator_optimizer = tf.keras.optimizers.Adam(lr = 0.0002, beta_1 = 0.5, beta_2 = 0.999 )
discriminator_optimizer = tf.keras.optimizers.Adam(lr = 0.0002, beta_1 = 0.5, beta_2 = 0.999 )

num_examples_to_generate = 25
latent_dim  = 100
# We will reuse this seed overtime to visualize progress
seed = tf.random.normal([num_examples_to_generate, latent_dim])

print(seed.dtype)

print(conditional_gen.input)

# Notice the use of `tf.function`
# This annotation causes the function to be "compiled".
@tf.function
def train_step(images,target):
    # noise vector sampled from normal distribution
    noise = tf.random.normal([target.shape[0], latent_dim])
    # Train Discriminator with real labels
    with tf.GradientTape() as disc_tape1:
        generated_images = conditional_gen([noise,target], training=True)

        
        real_output = conditional_discriminator([images,target], training=True)
        real_targets = tf.ones_like(real_output)
        disc_loss1 = discriminator_loss(real_targets, real_output)
        
    # gradient calculation for discriminator for real labels    
    gradients_of_disc1 = disc_tape1.gradient(disc_loss1, conditional_discriminator.trainable_variables)
    
    # parameters optimization for discriminator for real labels   
    discriminator_optimizer.apply_gradients(zip(gradients_of_disc1,\
    conditional_discriminator.trainable_variables))
    
    # Train Discriminator with fake labels
    with tf.GradientTape() as disc_tape2:
        fake_output = conditional_discriminator([generated_images,target], training=True)
        fake_targets = tf.zeros_like(fake_output)
        disc_loss2 = discriminator_loss(fake_targets, fake_output)
    # gradient calculation for discriminator for fake labels 
    gradients_of_disc2 = disc_tape2.gradient(disc_loss2, conditional_discriminator.trainable_variables)
    
    
    # parameters optimization for discriminator for fake labels        
    discriminator_optimizer.apply_gradients(zip(gradients_of_disc2,\
    conditional_discriminator.trainable_variables))
    
    # Train Generator with real labels
    with tf.GradientTape() as gen_tape:
        generated_images = conditional_gen([noise,target], training=True)
        fake_output = conditional_discriminator([generated_images,target], training=True)
        real_targets = tf.ones_like(fake_output)
        gen_loss = generator_loss(real_targets, fake_output)

    # gradient calculation for generator for real labels     
    gradients_of_gen = gen_tape.gradient(gen_loss, conditional_gen.trainable_variables)
    
    # parameters optimization for generator for real labels
    generator_optimizer.apply_gradients(zip(gradients_of_gen,\
    conditional_gen.trainable_variables))    

def train(dataset, epochs):
    for epoch in range(epochs):
        start = time.time()
        i = 0
        D_loss_list, G_loss_list = [], []
        for image_batch,target in dataset:
            i += 1
            train_step(image_batch,target)
        print(epoch)        
        display.clear_output(wait=True)
        generate_and_save_images(conditional_gen,
                              epoch + 1,
                              seed)

#         # Save the model every 15 epochs
#         if (epoch + 1) % 15 == 0:
#             checkpoint.save(file_prefix = checkpoint_prefix)

        conditional_gen.save_weights('fashion/training_weights/gen_'+ str(epoch)+'.h5')
        conditional_discriminator.save_weights('fashion/training_weights/disc_'+ str(epoch)+'.h5')    
        print ('Time for epoch {} is {} sec'.format(epoch + 1, time.time()-start))

    # Generate after the final epoch
    display.clear_output(wait=True)
    generate_and_save_images(conditional_gen,
                            epochs,
                            seed)

def label_gen(n_classes=10):
    lab = tf.random.uniform((1,), minval=0, maxval=10, dtype=tf.dtypes.int32, seed=None, name=None)
    return tf.repeat(lab, [25], axis=None, name=None)

# Create dictionary of target classes
label_dict = {
 0: 'T-shirt/top',
 1: 'Trouser',
 2: 'Pullover',
 3: 'Dress',
 4: 'Coat',
 5: 'Sandal',
 6: 'Shirt',
 7: 'Sneaker',
 8: 'Bag',
 9: 'Ankle boot',
}

def generate_and_save_images(model, epoch, test_input):
  # Notice `training` is set to False.
  # This is so all layers run in inference mode (batchnorm).
    labels = label_gen()
    predictions = model([test_input, labels], training=False)
    print(predictions.shape)
    fig = plt.figure(figsize=(4,4))

    print("Generated Images are Conditioned on Label:", label_dict[np.array(labels)[0]])
    for i in range(predictions.shape[0]):
        pred = (predictions[i, :, :, 0] + 1) * 127.5
        pred = np.array(pred)
        plt.subplot(5, 5, i+1)
        plt.imshow(pred.astype(np.uint8), cmap='gray')
        plt.axis('off')

    plt.savefig('fashion/images/image_at_epoch_{:d}.png'.format(epoch))
    plt.show()

train(train_dataset, 2)

conditional_gen.load_weights('fashion/training_weights/gen_1.h5')

# example of interpolating between generated faces

fig = plt.figure(figsize=(10,10))
# generate points in latent space as input for the generator
def generate_latent_points(latent_dim, n_samples, n_classes=10):
    # generate points in the latent space
    x_input = randn(latent_dim * n_samples)
    # reshape into a batch of inputs for the network
    z_input = x_input.reshape(n_samples, latent_dim)
    return z_input

# uniform interpolation between two points in latent space
def interpolate_points(p1, p2, n_steps=10):
    # interpolate ratios between the points
    ratios = linspace(0, 1, num=n_steps)
    # linear interpolate vectors
    vectors = list()
    for ratio in ratios:
        v = (1.0 - ratio) * p1 + ratio * p2
        vectors.append(v)
    return asarray(vectors)


# load model
pts = generate_latent_points(100, 2)
# interpolate points in latent space
interpolated = interpolate_points(pts[0], pts[1])
# generate images
from matplotlib import gridspec

output = None
for label in range(10):
    labels = tf.ones(10) * label
    predictions = conditional_gen([interpolated, labels], training=False)
    if output is None:
        output = predictions
    else:
        output = np.concatenate((output,predictions))

k = 0
nrow = 10
ncol = 10
fig = plt.figure(figsize=(15,15))
gs = gridspec.GridSpec(nrow, ncol, width_ratios=[1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
     wspace=0.0, hspace=0.0, top=0.95, bottom=0.05, left=0.17, right=0.845) 


for i in range(10):
    for j in range(10):
        pred = (output[k, :, :, :] + 1 ) * 127.5
        ax= plt.subplot(gs[i,j])
        pred = np.array(pred)
        ax.imshow(pred.astype(np.uint8), cmap='gray')
        ax.set_xticklabels([])
        ax.set_yticklabels([])
        ax.axis('off')
        k += 1   


plt.savefig('result.png',  dpi=300)
plt.show()

print(pred.shape)