StandardCAS-NSTID
commited on
Commit
•
065e0c7
1
Parent(s):
1766e29
Create Estallie_Trainer.py
Browse files- Estallie_Trainer.py +60 -0
Estallie_Trainer.py
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import tensorflow as tf
|
3 |
+
from tensorflow.keras.preprocessing.image import ImageDataGenerator
|
4 |
+
|
5 |
+
# Define constants
|
6 |
+
IMAGE_SIZE = (512, 512)
|
7 |
+
BATCH_SIZE = 4
|
8 |
+
EPOCHS = 10
|
9 |
+
TRAIN_DIR = 'T'
|
10 |
+
VALID_DIR = 'T'
|
11 |
+
MODEL_PATH = 'nsfw_classifier.h5'
|
12 |
+
|
13 |
+
# Create an image data generator for training data
|
14 |
+
train_datagen = ImageDataGenerator(rescale=1./255)
|
15 |
+
train_generator = train_datagen.flow_from_directory(
|
16 |
+
TRAIN_DIR,
|
17 |
+
target_size=IMAGE_SIZE,
|
18 |
+
batch_size=BATCH_SIZE,
|
19 |
+
class_mode='binary')
|
20 |
+
|
21 |
+
# Create an image data generator for validation data
|
22 |
+
valid_datagen = ImageDataGenerator(rescale=1./255)
|
23 |
+
valid_generator = valid_datagen.flow_from_directory(
|
24 |
+
VALID_DIR,
|
25 |
+
target_size=IMAGE_SIZE,
|
26 |
+
batch_size=BATCH_SIZE,
|
27 |
+
class_mode='binary')
|
28 |
+
|
29 |
+
# Check if the model already exists
|
30 |
+
if os.path.exists(MODEL_PATH):
|
31 |
+
print("Loading existing model")
|
32 |
+
model = tf.keras.models.load_model(MODEL_PATH)
|
33 |
+
else:
|
34 |
+
print("Creating new model")
|
35 |
+
# Define the model
|
36 |
+
model = tf.keras.models.Sequential([
|
37 |
+
tf.keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=(IMAGE_SIZE[0], IMAGE_SIZE[1], 3)),
|
38 |
+
tf.keras.layers.MaxPooling2D(2, 2),
|
39 |
+
tf.keras.layers.Conv2D(64, (3, 3), activation='relu'),
|
40 |
+
tf.keras.layers.MaxPooling2D(2, 2),
|
41 |
+
tf.keras.layers.Flatten(),
|
42 |
+
tf.keras.layers.Dense(512, activation='relu'),
|
43 |
+
tf.keras.layers.Dense(1, activation='sigmoid')
|
44 |
+
])
|
45 |
+
|
46 |
+
# Compile the model
|
47 |
+
model.compile(loss='binary_crossentropy',
|
48 |
+
optimizer='adam',
|
49 |
+
metrics=['accuracy'])
|
50 |
+
|
51 |
+
# Train the model
|
52 |
+
history = model.fit(
|
53 |
+
train_generator,
|
54 |
+
steps_per_epoch=train_generator.samples // BATCH_SIZE,
|
55 |
+
epochs=EPOCHS,
|
56 |
+
validation_data=valid_generator,
|
57 |
+
validation_steps=valid_generator.samples // BATCH_SIZE)
|
58 |
+
|
59 |
+
# Save the model
|
60 |
+
model.save(MODEL_PATH)
|