Question
Transcribed Text
HW3
β Train a discriminator/generator pair on CIFAR10 dataset
utilizing techniques from DCGAN and Wasserstein GANs
Baseline Model for DCGAN
β Generator
β noise_input = (100,);
β text_input = (119,);
β # num of (hair, eyes) pairs
β text_emb = Dense(256,βreluβ)(text_input);
β concatenate([noise_input, text_emb]);
β Dense(4*4*512); Reshape((4, 4, 512));
β Batchnorm(mom=0.9); Relu;
β Conv2DTranspose(256, kernel=5);
β Batchnorm(mom=0.9); Relu;
β Conv2DTranspose(128, kernel=5);
β Batchnorm(mom=0.9); Relu;
β Conv2DTranspose(64, kernel=5);
β Batchnorm(mom=0.9); Relu;
β Conv2DTranspose(3, kernel=5);
β Tanh;
β Discriminator
β image_input = (64,64,3);
β text_input = (119,);
β text_emb = Dense(256,βreluβ)(text_input);
β text_emb = Reshape((1,1,256))(text_emb);
β tiled_emb = tile(text_emb, [1,4,4,1]);
β Conv2D(64 ,kernel=5)(image_input); LeakyRelu;
β Conv2D(128, kernel=5);
β Batchnorm(mom=0.9); LeakyRelu;
β Conv2D(256, kernel=5);
β Batchnorm(mom=0.9); LeakyReLu;
β Conv2D(512, kernel=5);
β Batchnorm(mom=0.9);
β image_feat = LeakyRelu;
β concatenate([image_feat, tiled_emb]);
β Conv2D(512, kernel=1, strides=(1,1));
β Flatten;
β Dense(1, βsigmoidβ);
β Training
β Adam(lr = 0.0002, beta = 0.5)
Solution Preview
This material may consist of step-by-step explanations on how to solve a problem or examples of proper writing, including the use of citations, references, bibliographies, and formatting. This material is made available for the sole purpose of studying and learning - misuse is strictly forbidden.
rom math import floorfrom numpy import ones
from numpy import expand_dims
from numpy import log
from numpy import mean
from numpy import std
from numpy import exp
from numpy.random import shuffle
from keras.applications.inception_v3 import InceptionV3
from keras.applications.inception_v3 import preprocess_input
from keras.datasets import cifar10
from skimage.transform import resize
from numpy import asarray
# scale an array of images to a new size
def scale_images(images, new_shape):
images_list = list()
for image in images:
# resize with nearest neighbor interpolation
new_image = resize(image, new_shape, 0)
# store
images_list.append(new_image)
return asarray(images_list)
# assumes images have any shape and pixels in [0,255]
def calculate_inception_score(images, n_split=10, eps=1E-16):
# load inception v3 model
model = load('gen_model_020.h5')
# enumerate splits of images/predictions
scores = list()
n_part = floor(images.shape[0] / n_split)
for i in range(n_split):
# retrieve images
ix_start, ix_end = i * n_part, (i+1) * n_part
subset = images[ix_start:ix_end]
# convert from uint8 to float32
subset = subset.astype('float32')
# scale images to the required size
subset = scale_images(subset, (299,299,3))
# pre-process images, scale to [-1,1]
subset = preprocess_input(subset)
# predict p(y|x)
p_yx = model.predict(subset)
# calculate p(y)
p_y = expand_dims(p_yx.mean(axis=0), 0)
# calculate KL divergence using log probabilities
kl_d = p_yx * (log(p_yx + eps) - log(p_y + eps))
# sum over classes
sum_kl_d = kl_d.sum(axis=1)
# average over images
avg_kl_d = mean(sum_kl_d)
# undo the log
is_score = exp(avg_kl_d)
# store
scores.append(is_score)
# average across images
is_avg, is_std = mean(scores), std(scores)
return is_avg, is_std
# load cifar10 images
(images, _), (_, _) = cifar10.load_data()
# shuffle images
shuffle(images)...
By purchasing this solution you'll be able to access the following files:
Solution.zip.
Return to homework library