|
1
|
|
|
import sys |
|
2
|
|
|
import click |
|
3
|
|
|
import numpy as np |
|
4
|
|
|
|
|
5
|
|
|
from .disk_operations import Disk |
|
6
|
|
|
from .styling_observer import StylingObserver |
|
7
|
|
|
from .algorithm import NSTAlgorithm, AlogirthmParameters |
|
8
|
|
|
from .nst_tf_algorithm import NSTAlgorithmRunner |
|
9
|
|
|
from .termination_condition_adapter_factory import TerminationConditionAdapterFactory |
|
10
|
|
|
from .nst_image import ImageManager, noisy, convert_to_uint8 |
|
11
|
|
|
from .production_networks import NetworkDesign |
|
12
|
|
|
from .pretrained_model import ModelHandlerFacility |
|
13
|
|
|
|
|
14
|
|
|
|
|
15
|
|
|
def load_pretrained_model_functions(): |
|
16
|
|
|
# future work: discover dynamically the modules inside the pre_trained_model |
|
17
|
|
|
# package |
|
18
|
|
|
from .pre_trained_models import vgg |
|
19
|
|
|
return vgg |
|
20
|
|
|
|
|
21
|
|
|
|
|
22
|
|
|
def read_images(content, style): |
|
23
|
|
|
# todo dynamically find means |
|
24
|
|
|
means = np.array([123.68, 116.779, 103.939]).reshape((1,1,1,3)) # means |
|
25
|
|
|
|
|
26
|
|
|
image_manager = ImageManager.default(means) |
|
27
|
|
|
|
|
28
|
|
|
# probably load each image in separate thread and then join |
|
29
|
|
|
image_manager.load_from_disk(content, 'content') |
|
30
|
|
|
image_manager.load_from_disk(style, 'style') |
|
31
|
|
|
|
|
32
|
|
|
if not image_manager.images_compatible: |
|
33
|
|
|
print("Given CONTENT image '{content_image}' has 'height' x 'width' x " |
|
34
|
|
|
f"'color_channels': {image_manager.content_image.matrix.shape}") |
|
35
|
|
|
print("Given STYLE image '{style_image}' has 'height' x 'width' x " |
|
36
|
|
|
f"'color_channels': {image_manager.style_image.matrix.shape}") |
|
37
|
|
|
print('Expected to receive images (matrices) of identical shape') |
|
38
|
|
|
print('Exiting..') |
|
39
|
|
|
sys.exit(1) |
|
40
|
|
|
|
|
41
|
|
|
return image_manager.content_image, image_manager.style_image |
|
42
|
|
|
|
|
43
|
|
|
|
|
44
|
|
|
@click.command() |
|
45
|
|
|
@click.argument('content_image') |
|
46
|
|
|
@click.argument('style_image') |
|
47
|
|
|
@click.option('--iterations', '-it', type=int, default=100, show_default=True) |
|
48
|
|
|
@click.option('--location', '-l', type=str, default='.') |
|
49
|
|
|
def cli(content_image, style_image, iterations, location): |
|
50
|
|
|
|
|
51
|
|
|
termination_condition = 'max-iterations' |
|
52
|
|
|
|
|
53
|
|
|
content_image, style_image = read_images(content_image, style_image) |
|
54
|
|
|
|
|
55
|
|
|
load_pretrained_model_functions() |
|
56
|
|
|
model_design = type('ModelDesign', (), { |
|
57
|
|
|
'pretrained_model': ModelHandlerFacility.create('vgg'), |
|
58
|
|
|
'network_design': NetworkDesign.from_default_vgg() |
|
59
|
|
|
}) |
|
60
|
|
|
model_design.pretrained_model.load_model_layers() |
|
61
|
|
|
|
|
62
|
|
|
termination_condition = TerminationConditionAdapterFactory.create( |
|
63
|
|
|
termination_condition, |
|
64
|
|
|
iterations, |
|
65
|
|
|
) |
|
66
|
|
|
|
|
67
|
|
|
print(f' -- Termination Condition: {termination_condition.termination_condition}') |
|
68
|
|
|
|
|
69
|
|
|
algorithm_parameters = AlogirthmParameters( |
|
70
|
|
|
content_image, |
|
71
|
|
|
style_image, |
|
72
|
|
|
termination_condition, |
|
73
|
|
|
location, |
|
74
|
|
|
) |
|
75
|
|
|
|
|
76
|
|
|
algorithm = NSTAlgorithm(algorithm_parameters) |
|
77
|
|
|
|
|
78
|
|
|
noisy_ratio = 0.6 # ratio |
|
79
|
|
|
|
|
80
|
|
|
algorithm_runner = NSTAlgorithmRunner.default( |
|
81
|
|
|
lambda matrix: noisy(matrix, noisy_ratio), |
|
82
|
|
|
) |
|
83
|
|
|
|
|
84
|
|
|
algorithm_runner.progress_subject.add( |
|
85
|
|
|
termination_condition, |
|
86
|
|
|
) |
|
87
|
|
|
algorithm_runner.persistance_subject.add( |
|
88
|
|
|
StylingObserver(Disk.save_image, convert_to_uint8) |
|
89
|
|
|
) |
|
90
|
|
|
|
|
91
|
|
|
algorithm_runner.run(algorithm, model_design) |
|
92
|
|
|
|
boromir674 /
neural-style-transfer
