-
TensorFlow 기초 30-1 - 전이학습(기초 30 이어서)TensorFlow 2022. 12. 8. 16:42
# 전이학습이란? # 학습 데이터가 부족한 분야의 모델 구축을 위해 데이터가 풍부한 분야에서 훈련된 모델을 재사용하는 머신러닝 학습기법 # 구글이 만든 MobileVet V2 모델을 사용 # pip install tensorflow-datasets !pip install tensorflow-datasets # MobileVet V2 모델을 일부 재학습한 후 개/고양이 분류 모델을 생성 import os import numpy as np import matplotlib.pyplot as plt import tensorflow as tf import tensorflow_datasets as tfds # tfds.disable_progress_bar # 진행률 표시바를 보여주지 마! (raw_train, raw_validation, raw_test), metadata = tfds.load('cats_vs_dogs', split=['train[:80%]', 'train[80%:90%]', 'train[90%:]'], with_info=True, as_supervised=True) print(raw_train) print(raw_train.take(1)) # 한 개의 이지지와 해당 레이블을 읽기기 print(raw_validation) get_label_name = metadata.features['label'].int2str for image, label in raw_train.take(4): plt.figure() plt.imshow(image) plt.title(get_label_name(label)) plt.show() # image formating : MobileVet V2 모델이 원하기 때문에 IMG_SIZE = 160 def format_exam(image, label): # dataset 함수로 넣어주기 위함 image = tf.cast(image, tf.float32) image = (image / 127.5) -1 image = tf.image.resize(image, (IMG_SIZE, IMG_SIZE)) return image, label train = raw_train.map(format_exam) validation = raw_validation.map(format_exam) test = raw_test.map(format_exam) # 이미지 섞기 BATCH_SIZE = 32 SHUFFLE_BUFFER_SIZE = 1000 train_batchs = train.shuffle(SHUFFLE_BUFFER_SIZE).batch(BATCH_SIZE) # train만 shuffle 하기 validation_batchs = validation.batch(BATCH_SIZE) test_batchs = test.batch(BATCH_SIZE) for image_batch, label_batch in train_batchs.take(1): pass print(image_batch.shape) # (32, 160, 160, 3) # base model 설계(MobileNet V2) 설계 - 대량의 데이터로 학습을 끝낸 나이스한 분류 모델 IMG_SHAPE = (IMG_SIZE, IMG_SIZE, 3) base_model = tf.keras.applications.MobileNetV2(input_shape=IMG_SHAPE, include_top=False, weights='imagenet') feature_batch = base_model(image_batch) # 해당 이미지 특징 반환 print(feature_batch) # shape=(32, 5, 5, 1280) # 계층 동결 # 나이스한 모델 읽기가 끝남. - 합성곱 층 동결, 완전 연결층만 학습하는 방법을 사용 base_model.trainble = False # base_model은 학습시키지 않음 print(base_model.summary()) # 분류 모델링(설계) # base-model의 최종 출력 특징 : (None, 5, 5, 1280) <- 얘를 완전연결층에 맞도록 차원 축소, 즉 벡터화 작업이 필요 global_average_layer = tf.keras.layers.GlobalAveragePooling2D() # 공간 데이터에 대한 전역 평균 풀링 작업. feature를 1차원 벡터화. M feature_batch_average = global_average_layer(feature_batch) print(feature_batch_average.shape) # (32, 1280) prediction_layer = tf.keras.layers.Dense(1) prediction_batch = prediction_layer(feature_batch_average) print(prediction_batch.shape) model = tf.keras.Sequential([ base_model, # 특징 추출 베이스 모델 global_average_layer, # 출력값의 형태 변형을 위한 풀링 레이어 prediction_layer # 데이터 분석을 하는 완전 연결층층 ]) model.compile(optimizer = tf.keras.optimizers.RMSprop(learning_rate=0.0001), loss = tf.keras.losses.BinaryCrossentropy(from_logits=True), metrics=['accuracy']) print(model.summary()) # 학습 전 모델 성능 확인 vali_step = 20 loss0, accuracy0 = model.evaluate(validation_batchs, steps = vali_step) # step : 기본값은 None, 평가가 1회 완료되었을음 선언하기까지의 단계(샘플배치)의 총 갯수 print('학습 전 모델 loss : {:.2f}'.format(loss0)) print('학습 전 모델 accuracy : {:.2f}'.format(accuracy0)) # 학습 init_epochs = 10 history = model.fit(train_batchs, epochs=init_epochs, validation_data=validation_batchs) # 시각화 acc = history.history['accuracy'] val_acc = history.history['val_accuracy'] loss = history.history['loss'] val_loss = history.history['val_loss'] plt.figure(figsize=(8, 8)) plt.subplot(2, 1, 1) plt.plot(acc, label='train acc') plt.plot(val_acc, label = 'train validation acc') plt.legend(loc='lower right') plt.ylim([min(plt.ylim()), 1]) plt.subplot(2, 1, 2) plt.plot(acc, label='train loss') plt.plot(val_loss, label = 'train validation loss') plt.legend(loc='upper right') plt.ylim([0, 1.0]) plt.show() <console> Looking in indexes: https://pypi.org/simple, https://us-python.pkg.dev/colab-wheels/public/simple/ Requirement already satisfied: tensorflow-datasets in /usr/local/lib/python3.8/dist-packages (4.6.0) Requirement already satisfied: requests>=2.19.0 in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (2.23.0) Requirement already satisfied: protobuf>=3.12.2 in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (3.19.6) Requirement already satisfied: promise in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (2.3) Requirement already satisfied: termcolor in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (2.1.1) Requirement already satisfied: numpy in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (1.21.6) Requirement already satisfied: toml in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (0.10.2) Requirement already satisfied: tensorflow-metadata in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (1.11.0) Requirement already satisfied: tqdm in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (4.64.1) Requirement already satisfied: importlib-resources in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (5.10.0) Requirement already satisfied: absl-py in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (1.3.0) Requirement already satisfied: dill in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (0.3.6) Requirement already satisfied: etils[epath] in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (0.9.0) Requirement already satisfied: six in /usr/local/lib/python3.8/dist-packages (from tensorflow-datasets) (1.15.0) Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.8/dist-packages (from requests>=2.19.0->tensorflow-datasets) (2022.9.24) Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /usr/local/lib/python3.8/dist-packages (from requests>=2.19.0->tensorflow-datasets) (1.24.3) Requirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.8/dist-packages (from requests>=2.19.0->tensorflow-datasets) (3.0.4) Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.8/dist-packages (from requests>=2.19.0->tensorflow-datasets) (2.10) Requirement already satisfied: typing_extensions in /usr/local/lib/python3.8/dist-packages (from etils[epath]->tensorflow-datasets) (4.4.0) Requirement already satisfied: zipp in /usr/local/lib/python3.8/dist-packages (from etils[epath]->tensorflow-datasets) (3.11.0) Requirement already satisfied: googleapis-common-protos<2,>=1.52.0 in /usr/local/lib/python3.8/dist-packages (from tensorflow-metadata->tensorflow-datasets) (1.57.0) <PrefetchDataset element_spec=(TensorSpec(shape=(None, None, 3), dtype=tf.uint8, name=None), TensorSpec(shape=(), dtype=tf.int64, name=None))> <TakeDataset element_spec=(TensorSpec(shape=(None, None, 3), dtype=tf.uint8, name=None), TensorSpec(shape=(), dtype=tf.int64, name=None))> <PrefetchDataset element_spec=(TensorSpec(shape=(None, None, 3), dtype=tf.uint8, name=None), TensorSpec(shape=(), dtype=tf.int64, name=None))> (32, 160, 160, 3) Model: "mobilenetv2_1.00_160" __________________________________________________________________________________________________ Layer (type) Output Shape Param # Connected to ================================================================================================== input_2 (InputLayer) [(None, 160, 160, 3 0 [] )] Conv1 (Conv2D) (None, 80, 80, 32) 864 ['input_2[0][0]'] bn_Conv1 (BatchNormalization) (None, 80, 80, 32) 128 ['Conv1[0][0]'] Conv1_relu (ReLU) (None, 80, 80, 32) 0 ['bn_Conv1[0][0]'] expanded_conv_depthwise (Depth (None, 80, 80, 32) 288 ['Conv1_relu[0][0]'] wiseConv2D) expanded_conv_depthwise_BN (Ba (None, 80, 80, 32) 128 ['expanded_conv_depthwise[0][0]'] tchNormalization) expanded_conv_depthwise_relu ( (None, 80, 80, 32) 0 ['expanded_conv_depthwise_BN[0][0 ReLU) ]'] expanded_conv_project (Conv2D) (None, 80, 80, 16) 512 ['expanded_conv_depthwise_relu[0] [0]'] expanded_conv_project_BN (Batc (None, 80, 80, 16) 64 ['expanded_conv_project[0][0]'] hNormalization) block_1_expand (Conv2D) (None, 80, 80, 96) 1536 ['expanded_conv_project_BN[0][0]' ] block_1_expand_BN (BatchNormal (None, 80, 80, 96) 384 ['block_1_expand[0][0]'] ization) block_1_expand_relu (ReLU) (None, 80, 80, 96) 0 ['block_1_expand_BN[0][0]'] block_1_pad (ZeroPadding2D) (None, 81, 81, 96) 0 ['block_1_expand_relu[0][0]'] block_1_depthwise (DepthwiseCo (None, 40, 40, 96) 864 ['block_1_pad[0][0]'] nv2D) block_1_depthwise_BN (BatchNor (None, 40, 40, 96) 384 ['block_1_depthwise[0][0]'] malization) block_1_depthwise_relu (ReLU) (None, 40, 40, 96) 0 ['block_1_depthwise_BN[0][0]'] block_1_project (Conv2D) (None, 40, 40, 24) 2304 ['block_1_depthwise_relu[0][0]'] block_1_project_BN (BatchNorma (None, 40, 40, 24) 96 ['block_1_project[0][0]'] lization) block_2_expand (Conv2D) (None, 40, 40, 144) 3456 ['block_1_project_BN[0][0]'] block_2_expand_BN (BatchNormal (None, 40, 40, 144) 576 ['block_2_expand[0][0]'] ization) block_2_expand_relu (ReLU) (None, 40, 40, 144) 0 ['block_2_expand_BN[0][0]'] block_2_depthwise (DepthwiseCo (None, 40, 40, 144) 1296 ['block_2_expand_relu[0][0]'] nv2D) block_2_depthwise_BN (BatchNor (None, 40, 40, 144) 576 ['block_2_depthwise[0][0]'] malization) block_2_depthwise_relu (ReLU) (None, 40, 40, 144) 0 ['block_2_depthwise_BN[0][0]'] block_2_project (Conv2D) (None, 40, 40, 24) 3456 ['block_2_depthwise_relu[0][0]'] block_2_project_BN (BatchNorma (None, 40, 40, 24) 96 ['block_2_project[0][0]'] lization) block_2_add (Add) (None, 40, 40, 24) 0 ['block_1_project_BN[0][0]', 'block_2_project_BN[0][0]'] block_3_expand (Conv2D) (None, 40, 40, 144) 3456 ['block_2_add[0][0]'] block_3_expand_BN (BatchNormal (None, 40, 40, 144) 576 ['block_3_expand[0][0]'] ization) block_3_expand_relu (ReLU) (None, 40, 40, 144) 0 ['block_3_expand_BN[0][0]'] block_3_pad (ZeroPadding2D) (None, 41, 41, 144) 0 ['block_3_expand_relu[0][0]'] block_3_depthwise (DepthwiseCo (None, 20, 20, 144) 1296 ['block_3_pad[0][0]'] nv2D) block_3_depthwise_BN (BatchNor (None, 20, 20, 144) 576 ['block_3_depthwise[0][0]'] malization) block_3_depthwise_relu (ReLU) (None, 20, 20, 144) 0 ['block_3_depthwise_BN[0][0]'] block_3_project (Conv2D) (None, 20, 20, 32) 4608 ['block_3_depthwise_relu[0][0]'] block_3_project_BN (BatchNorma (None, 20, 20, 32) 128 ['block_3_project[0][0]'] lization) block_4_expand (Conv2D) (None, 20, 20, 192) 6144 ['block_3_project_BN[0][0]'] block_4_expand_BN (BatchNormal (None, 20, 20, 192) 768 ['block_4_expand[0][0]'] ization) block_4_expand_relu (ReLU) (None, 20, 20, 192) 0 ['block_4_expand_BN[0][0]'] block_4_depthwise (DepthwiseCo (None, 20, 20, 192) 1728 ['block_4_expand_relu[0][0]'] nv2D) block_4_depthwise_BN (BatchNor (None, 20, 20, 192) 768 ['block_4_depthwise[0][0]'] malization) block_4_depthwise_relu (ReLU) (None, 20, 20, 192) 0 ['block_4_depthwise_BN[0][0]'] block_4_project (Conv2D) (None, 20, 20, 32) 6144 ['block_4_depthwise_relu[0][0]'] block_4_project_BN (BatchNorma (None, 20, 20, 32) 128 ['block_4_project[0][0]'] lization) block_4_add (Add) (None, 20, 20, 32) 0 ['block_3_project_BN[0][0]', 'block_4_project_BN[0][0]'] block_5_expand (Conv2D) (None, 20, 20, 192) 6144 ['block_4_add[0][0]'] block_5_expand_BN (BatchNormal (None, 20, 20, 192) 768 ['block_5_expand[0][0]'] ization) block_5_expand_relu (ReLU) (None, 20, 20, 192) 0 ['block_5_expand_BN[0][0]'] block_5_depthwise (DepthwiseCo (None, 20, 20, 192) 1728 ['block_5_expand_relu[0][0]'] nv2D) block_5_depthwise_BN (BatchNor (None, 20, 20, 192) 768 ['block_5_depthwise[0][0]'] malization) block_5_depthwise_relu (ReLU) (None, 20, 20, 192) 0 ['block_5_depthwise_BN[0][0]'] block_5_project (Conv2D) (None, 20, 20, 32) 6144 ['block_5_depthwise_relu[0][0]'] block_5_project_BN (BatchNorma (None, 20, 20, 32) 128 ['block_5_project[0][0]'] lization) block_5_add (Add) (None, 20, 20, 32) 0 ['block_4_add[0][0]', 'block_5_project_BN[0][0]'] block_6_expand (Conv2D) (None, 20, 20, 192) 6144 ['block_5_add[0][0]'] block_6_expand_BN (BatchNormal (None, 20, 20, 192) 768 ['block_6_expand[0][0]'] ization) block_6_expand_relu (ReLU) (None, 20, 20, 192) 0 ['block_6_expand_BN[0][0]'] block_6_pad (ZeroPadding2D) (None, 21, 21, 192) 0 ['block_6_expand_relu[0][0]'] block_6_depthwise (DepthwiseCo (None, 10, 10, 192) 1728 ['block_6_pad[0][0]'] nv2D) block_6_depthwise_BN (BatchNor (None, 10, 10, 192) 768 ['block_6_depthwise[0][0]'] malization) block_6_depthwise_relu (ReLU) (None, 10, 10, 192) 0 ['block_6_depthwise_BN[0][0]'] block_6_project (Conv2D) (None, 10, 10, 64) 12288 ['block_6_depthwise_relu[0][0]'] block_6_project_BN (BatchNorma (None, 10, 10, 64) 256 ['block_6_project[0][0]'] lization) block_7_expand (Conv2D) (None, 10, 10, 384) 24576 ['block_6_project_BN[0][0]'] block_7_expand_BN (BatchNormal (None, 10, 10, 384) 1536 ['block_7_expand[0][0]'] ization) block_7_expand_relu (ReLU) (None, 10, 10, 384) 0 ['block_7_expand_BN[0][0]'] block_7_depthwise (DepthwiseCo (None, 10, 10, 384) 3456 ['block_7_expand_relu[0][0]'] nv2D) block_7_depthwise_BN (BatchNor (None, 10, 10, 384) 1536 ['block_7_depthwise[0][0]'] malization) block_7_depthwise_relu (ReLU) (None, 10, 10, 384) 0 ['block_7_depthwise_BN[0][0]'] block_7_project (Conv2D) (None, 10, 10, 64) 24576 ['block_7_depthwise_relu[0][0]'] block_7_project_BN (BatchNorma (None, 10, 10, 64) 256 ['block_7_project[0][0]'] lization) block_7_add (Add) (None, 10, 10, 64) 0 ['block_6_project_BN[0][0]', 'block_7_project_BN[0][0]'] block_8_expand (Conv2D) (None, 10, 10, 384) 24576 ['block_7_add[0][0]'] block_8_expand_BN (BatchNormal (None, 10, 10, 384) 1536 ['block_8_expand[0][0]'] ization) block_8_expand_relu (ReLU) (None, 10, 10, 384) 0 ['block_8_expand_BN[0][0]'] block_8_depthwise (DepthwiseCo (None, 10, 10, 384) 3456 ['block_8_expand_relu[0][0]'] nv2D) block_8_depthwise_BN (BatchNor (None, 10, 10, 384) 1536 ['block_8_depthwise[0][0]'] malization) block_8_depthwise_relu (ReLU) (None, 10, 10, 384) 0 ['block_8_depthwise_BN[0][0]'] block_8_project (Conv2D) (None, 10, 10, 64) 24576 ['block_8_depthwise_relu[0][0]'] block_8_project_BN (BatchNorma (None, 10, 10, 64) 256 ['block_8_project[0][0]'] lization) block_8_add (Add) (None, 10, 10, 64) 0 ['block_7_add[0][0]', 'block_8_project_BN[0][0]'] block_9_expand (Conv2D) (None, 10, 10, 384) 24576 ['block_8_add[0][0]'] block_9_expand_BN (BatchNormal (None, 10, 10, 384) 1536 ['block_9_expand[0][0]'] ization) block_9_expand_relu (ReLU) (None, 10, 10, 384) 0 ['block_9_expand_BN[0][0]'] block_9_depthwise (DepthwiseCo (None, 10, 10, 384) 3456 ['block_9_expand_relu[0][0]'] nv2D) block_9_depthwise_BN (BatchNor (None, 10, 10, 384) 1536 ['block_9_depthwise[0][0]'] malization) block_9_depthwise_relu (ReLU) (None, 10, 10, 384) 0 ['block_9_depthwise_BN[0][0]'] block_9_project (Conv2D) (None, 10, 10, 64) 24576 ['block_9_depthwise_relu[0][0]'] block_9_project_BN (BatchNorma (None, 10, 10, 64) 256 ['block_9_project[0][0]'] lization) block_9_add (Add) (None, 10, 10, 64) 0 ['block_8_add[0][0]', 'block_9_project_BN[0][0]'] block_10_expand (Conv2D) (None, 10, 10, 384) 24576 ['block_9_add[0][0]'] block_10_expand_BN (BatchNorma (None, 10, 10, 384) 1536 ['block_10_expand[0][0]'] lization) block_10_expand_relu (ReLU) (None, 10, 10, 384) 0 ['block_10_expand_BN[0][0]'] block_10_depthwise (DepthwiseC (None, 10, 10, 384) 3456 ['block_10_expand_relu[0][0]'] onv2D) block_10_depthwise_BN (BatchNo (None, 10, 10, 384) 1536 ['block_10_depthwise[0][0]'] rmalization) block_10_depthwise_relu (ReLU) (None, 10, 10, 384) 0 ['block_10_depthwise_BN[0][0]'] block_10_project (Conv2D) (None, 10, 10, 96) 36864 ['block_10_depthwise_relu[0][0]'] block_10_project_BN (BatchNorm (None, 10, 10, 96) 384 ['block_10_project[0][0]'] alization) block_11_expand (Conv2D) (None, 10, 10, 576) 55296 ['block_10_project_BN[0][0]'] block_11_expand_BN (BatchNorma (None, 10, 10, 576) 2304 ['block_11_expand[0][0]'] lization) block_11_expand_relu (ReLU) (None, 10, 10, 576) 0 ['block_11_expand_BN[0][0]'] block_11_depthwise (DepthwiseC (None, 10, 10, 576) 5184 ['block_11_expand_relu[0][0]'] onv2D) block_11_depthwise_BN (BatchNo (None, 10, 10, 576) 2304 ['block_11_depthwise[0][0]'] rmalization) block_11_depthwise_relu (ReLU) (None, 10, 10, 576) 0 ['block_11_depthwise_BN[0][0]'] block_11_project (Conv2D) (None, 10, 10, 96) 55296 ['block_11_depthwise_relu[0][0]'] block_11_project_BN (BatchNorm (None, 10, 10, 96) 384 ['block_11_project[0][0]'] alization) block_11_add (Add) (None, 10, 10, 96) 0 ['block_10_project_BN[0][0]', 'block_11_project_BN[0][0]'] block_12_expand (Conv2D) (None, 10, 10, 576) 55296 ['block_11_add[0][0]'] block_12_expand_BN (BatchNorma (None, 10, 10, 576) 2304 ['block_12_expand[0][0]'] lization) block_12_expand_relu (ReLU) (None, 10, 10, 576) 0 ['block_12_expand_BN[0][0]'] block_12_depthwise (DepthwiseC (None, 10, 10, 576) 5184 ['block_12_expand_relu[0][0]'] onv2D) block_12_depthwise_BN (BatchNo (None, 10, 10, 576) 2304 ['block_12_depthwise[0][0]'] rmalization) block_12_depthwise_relu (ReLU) (None, 10, 10, 576) 0 ['block_12_depthwise_BN[0][0]'] block_12_project (Conv2D) (None, 10, 10, 96) 55296 ['block_12_depthwise_relu[0][0]'] block_12_project_BN (BatchNorm (None, 10, 10, 96) 384 ['block_12_project[0][0]'] alization) block_12_add (Add) (None, 10, 10, 96) 0 ['block_11_add[0][0]', 'block_12_project_BN[0][0]'] block_13_expand (Conv2D) (None, 10, 10, 576) 55296 ['block_12_add[0][0]'] block_13_expand_BN (BatchNorma (None, 10, 10, 576) 2304 ['block_13_expand[0][0]'] lization) block_13_expand_relu (ReLU) (None, 10, 10, 576) 0 ['block_13_expand_BN[0][0]'] block_13_pad (ZeroPadding2D) (None, 11, 11, 576) 0 ['block_13_expand_relu[0][0]'] block_13_depthwise (DepthwiseC (None, 5, 5, 576) 5184 ['block_13_pad[0][0]'] onv2D) block_13_depthwise_BN (BatchNo (None, 5, 5, 576) 2304 ['block_13_depthwise[0][0]'] rmalization) block_13_depthwise_relu (ReLU) (None, 5, 5, 576) 0 ['block_13_depthwise_BN[0][0]'] block_13_project (Conv2D) (None, 5, 5, 160) 92160 ['block_13_depthwise_relu[0][0]'] block_13_project_BN (BatchNorm (None, 5, 5, 160) 640 ['block_13_project[0][0]'] alization) block_14_expand (Conv2D) (None, 5, 5, 960) 153600 ['block_13_project_BN[0][0]'] block_14_expand_BN (BatchNorma (None, 5, 5, 960) 3840 ['block_14_expand[0][0]'] lization) block_14_expand_relu (ReLU) (None, 5, 5, 960) 0 ['block_14_expand_BN[0][0]'] block_14_depthwise (DepthwiseC (None, 5, 5, 960) 8640 ['block_14_expand_relu[0][0]'] onv2D) block_14_depthwise_BN (BatchNo (None, 5, 5, 960) 3840 ['block_14_depthwise[0][0]'] rmalization) block_14_depthwise_relu (ReLU) (None, 5, 5, 960) 0 ['block_14_depthwise_BN[0][0]'] block_14_project (Conv2D) (None, 5, 5, 160) 153600 ['block_14_depthwise_relu[0][0]'] block_14_project_BN (BatchNorm (None, 5, 5, 160) 640 ['block_14_project[0][0]'] alization) block_14_add (Add) (None, 5, 5, 160) 0 ['block_13_project_BN[0][0]', 'block_14_project_BN[0][0]'] block_15_expand (Conv2D) (None, 5, 5, 960) 153600 ['block_14_add[0][0]'] block_15_expand_BN (BatchNorma (None, 5, 5, 960) 3840 ['block_15_expand[0][0]'] lization) block_15_expand_relu (ReLU) (None, 5, 5, 960) 0 ['block_15_expand_BN[0][0]'] block_15_depthwise (DepthwiseC (None, 5, 5, 960) 8640 ['block_15_expand_relu[0][0]'] onv2D) block_15_depthwise_BN (BatchNo (None, 5, 5, 960) 3840 ['block_15_depthwise[0][0]'] rmalization) block_15_depthwise_relu (ReLU) (None, 5, 5, 960) 0 ['block_15_depthwise_BN[0][0]'] block_15_project (Conv2D) (None, 5, 5, 160) 153600 ['block_15_depthwise_relu[0][0]'] block_15_project_BN (BatchNorm (None, 5, 5, 160) 640 ['block_15_project[0][0]'] alization) block_15_add (Add) (None, 5, 5, 160) 0 ['block_14_add[0][0]', 'block_15_project_BN[0][0]'] block_16_expand (Conv2D) (None, 5, 5, 960) 153600 ['block_15_add[0][0]'] block_16_expand_BN (BatchNorma (None, 5, 5, 960) 3840 ['block_16_expand[0][0]'] lization) block_16_expand_relu (ReLU) (None, 5, 5, 960) 0 ['block_16_expand_BN[0][0]'] block_16_depthwise (DepthwiseC (None, 5, 5, 960) 8640 ['block_16_expand_relu[0][0]'] onv2D) block_16_depthwise_BN (BatchNo (None, 5, 5, 960) 3840 ['block_16_depthwise[0][0]'] rmalization) block_16_depthwise_relu (ReLU) (None, 5, 5, 960) 0 ['block_16_depthwise_BN[0][0]'] block_16_project (Conv2D) (None, 5, 5, 320) 307200 ['block_16_depthwise_relu[0][0]'] block_16_project_BN (BatchNorm (None, 5, 5, 320) 1280 ['block_16_project[0][0]'] alization) Conv_1 (Conv2D) (None, 5, 5, 1280) 409600 ['block_16_project_BN[0][0]'] Conv_1_bn (BatchNormalization) (None, 5, 5, 1280) 5120 ['Conv_1[0][0]'] out_relu (ReLU) (None, 5, 5, 1280) 0 ['Conv_1_bn[0][0]'] ================================================================================================== Total params: 2,257,984 Trainable params: 2,223,872 Non-trainable params: 34,112 __________________________________________________________________________________________________ None (32, 1280) (32, 1) Model: "sequential_1" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= mobilenetv2_1.00_160 (Funct (None, 5, 5, 1280) 2257984 ional) global_average_pooling2d_2 (None, 1280) 0 (GlobalAveragePooling2D) dense_1 (Dense) (None, 1) 1281 ================================================================= Total params: 2,259,265 Trainable params: 2,225,153 Non-trainable params: 34,112 _________________________________________________________________ None 20/20 [==============================] - 2s 93ms/step - loss: 0.8105 - accuracy: 0.4875 학습 전 모델 loss : 0.81 학습 전 모델 accuracy : 0.49 Epoch 1/10 582/582 [==============================] - 108s 167ms/step - loss: 0.0813 - accuracy: 0.9676 - val_loss: 0.0704 - val_accuracy: 0.9798 Epoch 2/10 582/582 [==============================] - 98s 166ms/step - loss: 0.0265 - accuracy: 0.9897 - val_loss: 0.1024 - val_accuracy: 0.9781 Epoch 3/10 582/582 [==============================] - 97s 164ms/step - loss: 0.0157 - accuracy: 0.9944 - val_loss: 0.0867 - val_accuracy: 0.9875 Epoch 4/10 582/582 [==============================] - 97s 165ms/step - loss: 0.0114 - accuracy: 0.9959 - val_loss: 0.1153 - val_accuracy: 0.9811 Epoch 5/10 582/582 [==============================] - 97s 165ms/step - loss: 0.0071 - accuracy: 0.9978 - val_loss: 0.1639 - val_accuracy: 0.9755 Epoch 6/10 582/582 [==============================] - 98s 166ms/step - loss: 0.0068 - accuracy: 0.9978 - val_loss: 0.0986 - val_accuracy: 0.9893 Epoch 7/10 582/582 [==============================] - 97s 165ms/step - loss: 0.0044 - accuracy: 0.9988 - val_loss: 0.1153 - val_accuracy: 0.9845 Epoch 8/10 582/582 [==============================] - 104s 177ms/step - loss: 0.0043 - accuracy: 0.9983 - val_loss: 0.1739 - val_accuracy: 0.9781 Epoch 9/10 582/582 [==============================] - 103s 172ms/step - loss: 0.0043 - accuracy: 0.9985 - val_loss: 0.1700 - val_accuracy: 0.9845 Epoch 10/10 582/582 [==============================] - 117s 199ms/step - loss: 0.0032 - accuracy: 0.9989 - val_loss: 0.1295 - val_accuracy: 0.9824사진 위에 dog, cat을 나눠보았다.
'TensorFlow' 카테고리의 다른 글
TensorFlow 기초 32 - 형태소 분석 후 word2vec을 이용하여 단어 간 유사도 확인 (0) 2022.12.09 TensorFlow 기초 31 - 워드 임베딩 vs 원핫 인코딩, 밀집표현, 단어 간 유사도 (1) 2022.12.09 TensorFlow 기초 30 - cnn을 통한 댕댕이와 냥이 분류 모델 (0) 2022.12.08 TensorFlow 기초 29 - CNN을 활용해 이미지 특징을 뽑아 Dense로 학습(컬러 사) (0) 2022.12.08 TensorFlow 기초 28 - 이미지 보강 - 이미지가 부족한 경우 기존 이미지를 변형시켜 이미지 수를 늘림 (1) 2022.12.07