65 lines
1.9 KiB
Python
65 lines
1.9 KiB
Python
#!/usr/bin/python3
|
|
|
|
from tensorflow.keras.preprocessing.image import ImageDataGenerator
|
|
|
|
img_size = (150, 150)
|
|
batch_size = 192
|
|
extract_path="../ds/raw-img"
|
|
|
|
datagen = ImageDataGenerator(
|
|
rescale=1.0/255,
|
|
validation_split=0.2
|
|
)
|
|
|
|
def __dg(subset):
|
|
return datagen.flow_from_directory(extract_path,
|
|
target_size = img_size,
|
|
batch_size = batch_size,
|
|
class_mode = "categorical",
|
|
subset = subset,
|
|
shuffle = True)
|
|
|
|
train_generator = __dg("training")
|
|
val_generator = __dg("validation")
|
|
|
|
|
|
from tensorflow.keras import models as m
|
|
from tensorflow.keras import layers as l
|
|
from tensorflow.keras import optimizers as o
|
|
|
|
model = m.Sequential([
|
|
l.Input(shape=(150, 150, 3)),
|
|
l.Conv2D(96, (11, 11), strides=4, activation='relu'),
|
|
l.BatchNormalization(),
|
|
l.MaxPooling2D((3, 3), strides=2),
|
|
|
|
l.Conv2D(192, (5, 5), activation='relu', padding='same'),
|
|
l.BatchNormalization(),
|
|
l.MaxPooling2D((3, 3), strides=2),
|
|
l.Conv2D(256, (3, 3), activation='relu', padding='same'),
|
|
l.Conv2D(256, (3, 3), activation='relu', padding='same'),
|
|
l.Conv2D(160, (3, 3), activation='relu', padding='same'),
|
|
l.BatchNormalization(),
|
|
l.MaxPooling2D((3, 3), strides=2),
|
|
l.Flatten(),
|
|
l.Dense(1024, activation='relu'),
|
|
l.Dropout(0.5),
|
|
l.Dense(1024, activation='relu'),
|
|
l.Dropout(0.5),
|
|
l.Dense(10, activation='softmax'),
|
|
])
|
|
|
|
model.compile(optimizer = o.Adam(learning_rate = 0.0001),
|
|
loss = 'categorical_crossentropy',
|
|
metrics = ['accuracy'])
|
|
|
|
print(model.summary())
|
|
|
|
model.load_weights("w2.weights.h5")
|
|
|
|
model.fit(train_generator,
|
|
epochs = 10,
|
|
validation_data = val_generator)
|
|
|
|
model.save_weights("w2.weights.h5")
|