InteriorDesign/interior_design_renovation.py at main · AnnaLeites/InteriorDesign · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
# -*- coding: utf-8 -*-
"""interior_design_renovation.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1mbZS2x3KIDWGM9L41UTtpvewuMQ_R8dN

## Setting everything up
"""

#!pip install tensorflow
#!pip install tensorflow_hub

import tensorflow as tf
import tensorflow_hub as hub
import numpy
from google.colab.patches import cv2_imshow

#vgg19_url = "https://tfhub.dev/google/tf2-preview/vgg19/feature_vector/2"
#vgg19_model = hub.load(vgg19_url)

model = tf.keras.applications.vgg19.VGG19(
    include_top=True,
    weights='imagenet',
    input_tensor=None,
    input_shape=(224, 224, 3),
)

"""## Uploading pictures

Upload pictures of your apartment that needs a new design here:
"""

from google.colab import files
uploaded_r = files.upload()

"""Upload interior design that you like here:"""

from google.colab import files
uploaded_d = files.upload()

"""## Pre-processing of the images"""

#1
style_image = tf.keras.preprocessing.image.load_img("fancy.jpg", target_size=(224, 224))
style_image = tf.keras.preprocessing.image.img_to_array(style_image)
style_image = tf.keras.applications.vgg19.preprocess_input(style_image)
#2
# style_image_2 = tf.keras.preprocessing.image.load_img('despic2.jpg', target_size=(224, 224))
# style_image_2 = tf.keras.preprocessing.image.img_to_array(style_image_2)
# style_image_2 = tf.keras.applications.vgg19.preprocess_input(style_image_2)

#1
old_apartment_image = tf.keras.preprocessing.image.load_img("nf.jpg", target_size=(224, 224))
old_apartment_image = tf.keras.preprocessing.image.img_to_array(old_apartment_image)
old_apartment_image = tf.keras.applications.vgg19.preprocess_input(old_apartment_image)
# #2
# old_apartment_image_2 = tf.keras.preprocessing.image.load_img('apic2.jpg', target_size=(224, 224))
# old_apartment_image_2 = tf.keras.preprocessing.image.img_to_array(old_apartment_image_2)
# old_apartment_image_2 = tf.keras.applications.vgg19.preprocess_input(old_apartment_image_2)
# #3
# old_apartment_image_3 = tf.keras.preprocessing.image.load_img('apic3.jpg', target_size=(224, 224))
# old_apartment_image_3 = tf.keras.preprocessing.image.img_to_array(old_apartment_image_3)
# old_apartment_image_3 = tf.keras.applications.vgg19.preprocess_input(old_apartment_image_3)

combined_images = tf.stack([style_image, old_apartment_image], axis=0)
features = model(combined_images)
style_features = features[0]
content_features = features[1]

"""## Style transferring"""

def get_content_loss(content, generated):
    return tf.reduce_mean(tf.square(content - generated))

def get_style_loss(style, generated):
    return tf.reduce_mean(tf.square(style - generated))
loss_style = get_style_loss(style_features, content_features)

def total_variation_loss(image):
    x_deltas, y_deltas = image[:, :-1, :, :] - image[:, 1:, :, :], image[:, :, :-1, :] - image[:, :, 1:, :]
    return tf.reduce_sum(tf.abs(x_deltas)) + tf.reduce_sum(tf.abs(y_deltas))

def get_total_loss(style_loss, content_loss, total_variation_weight=1e-9):
    return style_loss + content_loss + total_variation_weight * total_variation_loss(generated_image)

optimizer = tf.optimizers.Adam(learning_rate=0.002)

num_iterations = 100000

# #style_features = model(style_image_1, style_image_2)
# style_features = model(style_image_1)
# content_features_1 = model(old_apartment_image_1)
# content_features_2 = model(old_apartment_image_2)
# content_features_3 = model(old_apartment_image_3)

initial_image = tf.reshape(old_apartment_image, (1, 224, 224, 3))
#generated_image = tf.Variable(tf.random.uniform(shape=(1, 224, 224, 3), minval=0, maxval=255), trainable=True)
generated_image = tf.Variable(initial_image, trainable=True)


generated_features = model(generated_image)

for i in range(num_iterations):
    with tf.GradientTape() as tape:
        loss_content = get_content_loss(content_features, generated_features)
        loss_style = get_style_loss(style_features, generated_features)
        total_loss = get_total_loss(loss_style, loss_content)

    gradients = tape.gradient(total_loss, generated_image)
    optimizer.apply_gradients([(gradients, generated_image)])

    generated_image.assign(tf.clip_by_value(generated_image, 0, 255))

# content_weight = 1e3
# style_weight = 1e-2

# def get_content_loss(content, target):
#     return tf.reduce_mean(tf.square(content - target))

# def gram_matrix(input_tensor):
#     result = tf.linalg.einsum('bijc,bijd->bcd', input_tensor, input_tensor)
#     input_shape = tf.shape(input_tensor)
#     num_locations = tf.cast(input_shape[1]*input_shape[2], tf.float32)
#     return result / num_locations

# def get_style_loss(style, generated):
#     style_loss = tf.constant(0.0, dtype=tf.float32)
#     for style_feat, generated_feat in zip(style, generated):
#         gram_style = tf.linalg.einsum('bijc,bijd->bcd', style_feat, style_feat)
#         gram_generated = tf.linalg.einsum('bijc,bijd->bcd', generated_feat, generated_feat)
#         style_loss += tf.reduce_mean(tf.square(gram_style - gram_generated))
#     return style_loss

# generated_image_1 = tf.Variable(old_apartment_image_1, trainable=True)
# generated_image_2 = tf.Variable(old_apartment_image_2, trainable=True)
# generated_image_3 = tf.Variable(old_apartment_image_3, trainable=True)

# optimizer = tf.optimizers.Adam(learning_rate=0.02)

# num_iterations = 1000

# for i in range(num_iterations):
#     with tf.GradientTape() as tape:
#         generated_features_1 = model(generated_image_1)
#         loss_content = get_content_loss(content_features_1, generated_features_1)
#         loss_style = get_style_loss(style_features, generated_features_1)

#         total_loss = content_weight * loss_content + style_weight * loss_style

#     gradients = tape.gradient(total_loss, generated_image_1)
#     optimizer.apply_gradients([(gradients, generated_image_1)])

#     generated_image_1.assign(tf.clip_by_value(generated_image_1, 0, 255))

# tf.keras.preprocessing.image.save_img('generated_image_1.jpg', generated_image_1.numpy())
# cv2_imshow('generated_image_1.jpg')

generated_image = tf.cast(generated_image, tf.uint8)  # Convert to uint8 format
tf.keras.preprocessing.image.save_img('generated_image.jpg', generated_image[0])