[56일차] ASAC 빅데이터 분석가 4기(DL-8)

⚙️ Inception Net

📌 ImageNet으로 학습한 모델 가져오기

base_model = InceptionV3(weights='imagenet', include_top=False)

---

📌 기존 모델의 layer를 이용한 재학습

for layer in base_model.layers:
   layer.trainable = False
   
num_classes = len(os.listdir(train_data_sub_folder))

x = base_model.output

x = GlobalAveragePooling2D()(x)
x = Dropout(0.2)(x)
x = Dense(1024, activation="relu")(x)
x = Dropout(0.2)(x)
x = Dense(1024, activation="relu")(x)
x = Dropout(0.2)(x)
x = Dense(512, activation="relu")(x)

preds = Dense(num_classes, activation="softmax")(x)

model = Model(inputs = base_model.input, outputs = preds)

adam = optimizers.Adam(learning_rate=0.001, beta_1 = 0.9, beta_2 = 0.999)

model.compile(optimizer = adam, loss = "categorical_crossentropy", metrics = ["accuracy"])

train_datagen = ImageDataGenerator(
    
    validation_split=0.1,
    rescale = 1./255.,

    rotation_range=40,
    width_shift_range=0.2,
    height_shift_range=0.2,
    shear_range=0.2,
    fill_mode='nearest'
    )

valid_datagen=ImageDataGenerator(
    validation_split=0.1,
    rescale = 1./255.,
)

image_size = 299
batch_size = 1024

train_generator = train_datagen.flow_from_directory(
    train_data_sub_folder,
    subset = "training",
    target_size = (image_size, image_size),
    seed = 42, shuffle=True,
    batch_size = batch_size,
    class_mode = "categorical"
)

valid_generator = valid_datagen.flow_from_directory(
    train_data_sub_folder,
    subset = "training",
    target_size = (image_size, image_size),
    seed = 42, shuffle=False,
    batch_size = batch_size,
    class_mode = "categorical"
)

import os

cp_path = "training/cp-{epoch:04d}.ckpt"
cp_dir = os.path.dirname(cp_path)

cp_callback = tf.keras.callbacks.ModelCheckpoint(
    cp_path,
    verbose = 1,
    save_weights_only = True    
)

es_callback = tf.keras.callbacks.EarlyStopping(
    monitor = "val_loss",
    patience = 3
)

num_epochs = 50

my_steps_per_epoch = int(train_generator.n / batch_size)

history = model.fit_generator(
    
    train_generator,
    
    epochs = num_epochs,
    
    validation_data = valid_generator,

    steps_per_epoch = my_steps_per_epoch,

    use_multiprocessing = True,

    callbacks = [cp_callback, es_callback]

    )

---

📌 학습시킨 모델을 이용한 예측

test_set = []
test_set_ids = []
for curImage in os.listdir('/content/test'):
    test_set_ids.append(os.path.splitext(curImage)[0])
    curImage = cv2.imread('/content/test/'+curImage)
    test_set.append(cv2.resize(curImage,(image_size, image_size)))

test_set = np.array(test_set, np.float32)/255.0

predictions= model.predict(test_set)

classes= {index:breed for breed,index in train_generator.class_indices.items()}
column_names = [classes[i] for i in range(120)]

predictions_df = pd.DataFrame(predictions)
predictions_df.columns = column_names
predictions_df.insert(0,'id', test_set_ids)
predictions_df.set_index('id',inplace=True)

predictions_df.to_csv('final_submission.csv',sep=",")