def train(image_path, model_path, gpu, latent_width, color_channels, batch,
epoch, shape, kl_ratio, mode):
if not image_path[-1] == '/':
image_path += '/'
if not os.path.exists(os.path.dirname(image_path)):
os.makedirs(os.path.dirname(image_path))
if model_path[-1] == '/':
click.echo("Model Path should be a file path not a directory path. "
"Removing trailing '/' ...")
model_path = model_path[:-1]
file_paths = [
os.path.join(image_path, f) for f in os.listdir(image_path) if
os.path.isfile(os.path.join(image_path, f)) and not f.startswith('.')
]
vae = VAE(img_width=shape[0],
img_height=shape[1],
encode_layers=[1000, 700, 300],
decode_layers=[500, 700, 1000],
kl_ratio=kl_ratio,
flag_gpu=gpu,
latent_width=latent_width,
color_channels=color_channels,
mode=mode)
x_all = vae.load_images(file_paths)
vae.fit(x_all, batch_size=batch, n_epochs=epoch)
directory = os.path.dirname(model_path)
name = os.path.basename(model_path)
vae.save(directory, name)
def train(image_path, model_path, gpu, latent_width, color_channels, batch,
epoch, shape, kl_ratio, mode):
if not image_path[-1] == '/':
image_path += '/'
if not os.path.exists(os.path.dirname(image_path)):
os.makedirs(os.path.dirname(image_path))
if model_path[-1] == '/':
click.echo("Model Path should be a file path not a directory path. "
"Removing trailing '/' ...")
model_path = model_path[:-1]
file_paths = [os.path.join(image_path, f)
for f in os.listdir(image_path)
if os.path.isfile(os.path.join(image_path, f))
and not f.startswith('.')]
vae = VAE(img_width=shape[0],
img_height=shape[1],
encode_layers=[1000, 700, 300],
decode_layers=[500, 700, 1000],
kl_ratio=kl_ratio,
flag_gpu=gpu,
latent_width=latent_width,
color_channels=color_channels,
mode=mode)
x_all = vae.load_images(file_paths)
vae.fit(x_all, batch_size=batch, n_epochs=epoch)
directory = os.path.dirname(model_path)
name = os.path.basename(model_path)
vae.save(directory, name)
def generate(model_path, optimizer_path, meta_path, img_dir, number, extremity,
mean, format, image_multiplier):
click.echo("Loading Model...")
vae = VAE.load(model_path, optimizer_path, meta_path)
click.echo("Model Loaded")
click.echo("Saving Images to {0} as {1}".format(img_dir, format))
if not img_dir[-1] == '/':
img_dir += '/'
if not os.path.exists(os.path.dirname(img_dir)):
os.makedirs(os.path.dirname(img_dir))
variance = extremity
vec = np.random.normal(mean, variance,
(number, vae.latent_width)).astype('float32')
reconstructed = vae.inverse_transform(vec, test=True) * 255
for i in range(number):
im = reconstructed[i]
im = np.clip(image_multiplier * im, 0, 255)
im = Image.fromarray(np.squeeze(im.astype(np.uint8)))
fname = "{0}.{1}".format(i, format)
path = os.path.join(img_dir, fname)
im.save(path)
def generate(model_path, optimizer_path, meta_path, img_dir, number,
extremity, mean, format, image_multiplier):
click.echo("Loading Model...")
vae = VAE.load(model_path, optimizer_path, meta_path)
click.echo("Model Loaded")
click.echo("Saving Images to {0} as {1}".format(img_dir, format))
if not img_dir[-1] == '/':
img_dir += '/'
if not os.path.exists(os.path.dirname(img_dir)):
os.makedirs(os.path.dirname(img_dir))
variance = extremity
vec = np.random.normal(mean, variance,
(number, vae.latent_width)).astype('float32')
reconstructed = vae.inverse_transform(vec, test=True) * 255
for i in range(number):
im = reconstructed[i]
im = np.clip(image_multiplier*im, 0, 255)
im = Image.fromarray(np.squeeze(im.astype(np.uint8)))
fname = "{0}.{1}".format(i, format)
path = os.path.join(img_dir, fname)
im.save(path)