Upload 4 files
Browse files
ckpts/best_model.h5
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:a6d0361fa140c1dc3b279bcce8107c28b6e10a4e1bc31f770e5b071a44f5f76d
|
3 |
+
size 20800096
|
models/auto_encoder_gray2color.py
ADDED
@@ -0,0 +1,92 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
import tensorflow as tf
|
3 |
+
from tensorflow.keras.layers import Input, Conv2D, MaxPooling2D, UpSampling2D, BatchNormalization, Add, Concatenate, Multiply
|
4 |
+
from tensorflow.keras.models import Model
|
5 |
+
from tensorflow.keras.optimizers import Adam
|
6 |
+
|
7 |
+
# Spatial Attention Layer
|
8 |
+
# Define SpatialAttention layer
|
9 |
+
class SpatialAttention(tf.keras.layers.Layer):
|
10 |
+
def __init__(self, kernel_size=7, **kwargs):
|
11 |
+
super(SpatialAttention, self).__init__(**kwargs)
|
12 |
+
self.kernel_size = kernel_size
|
13 |
+
self.conv = Conv2D(filters=1, kernel_size=kernel_size, padding='same', activation='sigmoid')
|
14 |
+
|
15 |
+
def call(self, inputs):
|
16 |
+
avg_pool = tf.reduce_mean(inputs, axis=-1, keepdims=True)
|
17 |
+
max_pool = tf.reduce_max(inputs, axis=-1, keepdims=True)
|
18 |
+
concat = Concatenate()([avg_pool, max_pool])
|
19 |
+
attention = self.conv(concat)
|
20 |
+
return Multiply()([inputs, attention])
|
21 |
+
|
22 |
+
def get_config(self):
|
23 |
+
config = super(SpatialAttention, self).get_config()
|
24 |
+
config.update({'kernel_size': self.kernel_size})
|
25 |
+
return config
|
26 |
+
|
27 |
+
# Build Autoencoder
|
28 |
+
def build_autoencoder(height, width,):
|
29 |
+
input_img = Input(shape=(height, width, 1))
|
30 |
+
|
31 |
+
# Encoder
|
32 |
+
x = Conv2D(96, (3, 3), activation='relu', padding='same')(input_img)
|
33 |
+
x = BatchNormalization()(x)
|
34 |
+
x = SpatialAttention()(x)
|
35 |
+
x = MaxPooling2D((2, 2), padding='same')(x)
|
36 |
+
|
37 |
+
# Residual Block 1
|
38 |
+
residual = Conv2D(192, (1, 1), padding='same')(x)
|
39 |
+
x = Conv2D(192, (3, 3), activation='relu', padding='same')(x)
|
40 |
+
x = BatchNormalization()(x)
|
41 |
+
x = Conv2D(192, (3, 3), activation='relu', padding='same')(x)
|
42 |
+
x = BatchNormalization()(x)
|
43 |
+
x = Add()([x, residual])
|
44 |
+
x = SpatialAttention()(x)
|
45 |
+
x = MaxPooling2D((2, 2), padding='same')(x)
|
46 |
+
|
47 |
+
# Residual Block 2
|
48 |
+
residual = Conv2D(384, (1, 1), padding='same')(x)
|
49 |
+
x = Conv2D(384, (3, 3), activation='relu', padding='same')(x)
|
50 |
+
x = BatchNormalization()(x)
|
51 |
+
x = Conv2D(384, (3, 3), activation='relu', padding='same')(x)
|
52 |
+
x = BatchNormalization()(x)
|
53 |
+
x = Add()([x, residual])
|
54 |
+
x = SpatialAttention()(x)
|
55 |
+
encoded = MaxPooling2D((2, 2), padding='same')(x)
|
56 |
+
|
57 |
+
# Decoder
|
58 |
+
x = Conv2D(384, (3, 3), activation='relu', padding='same')(encoded)
|
59 |
+
x = BatchNormalization()(x)
|
60 |
+
x = SpatialAttention()(x)
|
61 |
+
x = UpSampling2D((2, 2))(x)
|
62 |
+
|
63 |
+
# Residual Block 3
|
64 |
+
residual = Conv2D(192, (1, 1), padding='same')(x)
|
65 |
+
x = Conv2D(192, (3, 3), activation='relu', padding='same')(x)
|
66 |
+
x = BatchNormalization()(x)
|
67 |
+
x = Conv2D(192, (3, 3), activation='relu', padding='same')(x)
|
68 |
+
x = BatchNormalization()(x)
|
69 |
+
x = Add()([x, residual])
|
70 |
+
x = SpatialAttention()(x)
|
71 |
+
x = UpSampling2D((2, 2))(x)
|
72 |
+
|
73 |
+
x = Conv2D(96, (3, 3), activation='relu', padding='same')(x)
|
74 |
+
x = BatchNormalization()(x)
|
75 |
+
x = SpatialAttention()(x)
|
76 |
+
x = UpSampling2D((2, 2))(x)
|
77 |
+
|
78 |
+
decoded = Conv2D(2, (3, 3), activation=None, padding='same')(x)
|
79 |
+
|
80 |
+
return Model(input_img, decoded)
|
81 |
+
|
82 |
+
|
83 |
+
|
84 |
+
|
85 |
+
|
86 |
+
if __name__ == "__main__":
|
87 |
+
# Define constants
|
88 |
+
HEIGHT, WIDTH = 512, 512
|
89 |
+
# Compile model
|
90 |
+
autoencoder = build_autoencoder()
|
91 |
+
autoencoder.summary()
|
92 |
+
autoencoder.compile(optimizer=Adam(learning_rate=7e-5), loss=tf.keras.losses.MeanSquaredError())
|
notebooks/autoencoder-grayscale-to-color-landscape.ipynb
ADDED
The diff for this file is too large to render.
See raw diff
|
|
requirements.txt
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
numpy==1.26.4
|
2 |
+
tensorflow==2.18.0
|
3 |
+
opencv-python==4.11.0.86
|
4 |
+
scikit-image==0.25.2
|
5 |
+
matplotlib==3.7.2
|