JosefJilek
commited on
Commit
•
40f30f3
1
Parent(s):
08c8253
v1.0
Browse files
main.py
ADDED
@@ -0,0 +1,158 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
|
3 |
+
import matplotlib.pyplot as plt
|
4 |
+
import numpy as np
|
5 |
+
import tensorflow as tf
|
6 |
+
|
7 |
+
from tensorflow import keras
|
8 |
+
from tensorflow.keras import layers
|
9 |
+
from tensorflow.keras.models import Sequential
|
10 |
+
|
11 |
+
import pathlib
|
12 |
+
|
13 |
+
from tensorflow.python.client import device_lib
|
14 |
+
print(device_lib.list_local_devices())
|
15 |
+
|
16 |
+
data_dir = "C:/Users/jilek/Downloads/AAT+"
|
17 |
+
data_dir = pathlib.Path(data_dir).with_suffix('')
|
18 |
+
|
19 |
+
data_dir_test = "C:/Users/jilek/Downloads/AAT+_TEST"
|
20 |
+
data_dir_test = pathlib.Path(data_dir_test).with_suffix('')
|
21 |
+
|
22 |
+
image_count = len(list(data_dir.glob('*/*.jpg')))
|
23 |
+
print(image_count)
|
24 |
+
|
25 |
+
batch_size = 1
|
26 |
+
img_height = 1024
|
27 |
+
img_width = 1024
|
28 |
+
|
29 |
+
train_ds = tf.keras.utils.image_dataset_from_directory(
|
30 |
+
data_dir,
|
31 |
+
validation_split=0.0,
|
32 |
+
#subset="training",
|
33 |
+
seed=123,
|
34 |
+
labels='inferred',
|
35 |
+
label_mode='categorical',
|
36 |
+
class_names=["C100", "C095", "C090", "C085", "C080", "C070", "C060", "C040", "C020"],
|
37 |
+
color_mode="grayscale", #grayscale
|
38 |
+
shuffle=True,
|
39 |
+
image_size=(img_height, img_width),
|
40 |
+
batch_size=batch_size)
|
41 |
+
|
42 |
+
val_ds = tf.keras.utils.image_dataset_from_directory(
|
43 |
+
data_dir_test,
|
44 |
+
validation_split=0.0,
|
45 |
+
#subset="validation",
|
46 |
+
seed=123,
|
47 |
+
labels='inferred',
|
48 |
+
label_mode='categorical',
|
49 |
+
class_names=["C100", "C095", "C090", "C085", "C080", "C070", "C060", "C040", "C020"],
|
50 |
+
color_mode="grayscale",
|
51 |
+
image_size=(img_height, img_width),
|
52 |
+
batch_size=batch_size)
|
53 |
+
|
54 |
+
class_names = train_ds.class_names
|
55 |
+
print(class_names)
|
56 |
+
|
57 |
+
for image_batch, labels_batch in train_ds:
|
58 |
+
print(image_batch.shape)
|
59 |
+
print(labels_batch.shape)
|
60 |
+
break
|
61 |
+
|
62 |
+
AUTOTUNE = tf.data.AUTOTUNE
|
63 |
+
|
64 |
+
data_augmentation = keras.Sequential(
|
65 |
+
[
|
66 |
+
layers.RandomFlip("horizontal_and_vertical",
|
67 |
+
input_shape=(img_height,
|
68 |
+
img_width,
|
69 |
+
1)), #rgb
|
70 |
+
#layers.RandomRotation(0.5),
|
71 |
+
#layers.RandomZoom(0.5),
|
72 |
+
]
|
73 |
+
)
|
74 |
+
|
75 |
+
train_ds = train_ds.shuffle(buffer_size=900).prefetch(buffer_size=AUTOTUNE) #.cache()
|
76 |
+
val_ds = val_ds.prefetch(buffer_size=AUTOTUNE) #.cache()
|
77 |
+
|
78 |
+
|
79 |
+
num_classes = len(class_names)
|
80 |
+
print(str(num_classes))
|
81 |
+
|
82 |
+
model = Sequential([
|
83 |
+
layers.Rescaling(1.0/255, input_shape=(img_height, img_width, 1)), #rgb
|
84 |
+
#layers.Dropout(0.0),
|
85 |
+
#layers.MaxPooling2D(pool_size=(8, 8)),
|
86 |
+
layers.Conv2D(4, (4, 4), strides=(2, 2), padding='valid', dilation_rate=(1, 1), groups=1, input_shape=(1024, 1024, 1), activation='relu'),
|
87 |
+
layers.Conv2D(8, (4, 4), strides=(2, 2), padding='valid', dilation_rate=(1, 1), groups=1, input_shape=(512, 512, 4), activation='relu'),
|
88 |
+
layers.Conv2D(16, (4, 4), strides=(4, 4), padding='valid', dilation_rate=(1, 1), groups=1, input_shape=(256, 256, 8), activation='relu'),
|
89 |
+
layers.Conv2D(32, (4, 4), strides=(4, 4), padding='valid', dilation_rate=(1, 1), groups=1, input_shape=(64, 64, 16), activation='relu'),
|
90 |
+
layers.Conv2D(64, (4, 4), strides=(4, 4), padding='valid', dilation_rate=(1, 1), groups=1, input_shape=(16, 16, 32), activation='relu'),
|
91 |
+
#layers.Conv2D(128, (4, 4), strides=(1, 1), padding='valid', dilation_rate=(1, 1), groups=1, input_shape=(8, 8, 64), activation='relu'),
|
92 |
+
#layers.Dropout(0.1),
|
93 |
+
layers.Flatten(),
|
94 |
+
layers.Dense(32, activation='relu'),
|
95 |
+
layers.Dense(num_classes, activation='softmax')
|
96 |
+
])
|
97 |
+
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-5),
|
98 |
+
loss=tf.keras.losses.CategoricalCrossentropy(),
|
99 |
+
metrics=['accuracy'])
|
100 |
+
|
101 |
+
model.summary()
|
102 |
+
model.save("./model/AAT+")
|
103 |
+
|
104 |
+
epochs = 130
|
105 |
+
history = model.fit(
|
106 |
+
train_ds,
|
107 |
+
validation_data=val_ds,
|
108 |
+
epochs=epochs
|
109 |
+
)
|
110 |
+
|
111 |
+
acc = history.history['accuracy']
|
112 |
+
val_acc = history.history['val_accuracy']
|
113 |
+
|
114 |
+
loss = history.history['loss']
|
115 |
+
val_loss = history.history['val_loss']
|
116 |
+
|
117 |
+
epochs_range = range(epochs)
|
118 |
+
|
119 |
+
plt.figure(figsize=(8, 8))
|
120 |
+
plt.subplot(1, 2, 1)
|
121 |
+
plt.plot(epochs_range, acc, label='Training Accuracy')
|
122 |
+
plt.plot(epochs_range, val_acc, label='Validation Accuracy')
|
123 |
+
plt.legend(loc='lower right')
|
124 |
+
plt.title('Training and Validation Accuracy')
|
125 |
+
|
126 |
+
plt.subplot(1, 2, 2)
|
127 |
+
plt.plot(epochs_range, loss, label='Training Loss')
|
128 |
+
plt.plot(epochs_range, val_loss, label='Validation Loss')
|
129 |
+
plt.legend(loc='upper right')
|
130 |
+
plt.title('Training and Validation Loss')
|
131 |
+
plt.show()
|
132 |
+
|
133 |
+
test_dir = "C:/Users/jilek/Downloads/AAT_T/"
|
134 |
+
for file_name in os.listdir(test_dir):
|
135 |
+
file_path = os.path.join(test_dir, file_name)
|
136 |
+
img = tf.keras.utils.load_img(
|
137 |
+
file_path, target_size=(img_height, img_width), color_mode="grayscale" #grayscale
|
138 |
+
)
|
139 |
+
img_array = tf.keras.utils.img_to_array(img)
|
140 |
+
img_array = tf.expand_dims(img_array, 0) # Create a batch
|
141 |
+
|
142 |
+
predictions = model.predict(img_array)
|
143 |
+
score = tf.nn.softmax(predictions[0])
|
144 |
+
|
145 |
+
print(file_name)
|
146 |
+
print(
|
147 |
+
"This image most likely belongs to {} with a {:.2f} percent confidence."
|
148 |
+
.format(class_names[np.argmax(score)], 100 * np.max(score))
|
149 |
+
)
|
150 |
+
|
151 |
+
|
152 |
+
|
153 |
+
|
154 |
+
|
155 |
+
|
156 |
+
|
157 |
+
|
158 |
+
|
model/AAT+/keras_metadata.pb
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:429c89709452f69986c9cc2c9549be3e6e6525d467876355348d373241545d32
|
3 |
+
size 23711
|
model/AAT+/saved_model.pb
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:bf5d62d94e882bf080ea5aa8f5ba9c4635bddd6a2cfd03d4be9ba875f2b5fa1a
|
3 |
+
size 148529
|
model/AAT+/variables/variables.data-00000-of-00001
ADDED
Binary file (258 kB). View file
|
|
model/AAT+/variables/variables.index
ADDED
Binary file (1.26 kB). View file
|
|