개와 고양이 분류를, 이미 잘 만들어진 뉴럴네트워크를 활용하여, 성능을 올려보자.¶
Stage 1: Install dependencies and setting up GPU environment¶
In [ ]:
# !pip install tensorflow-gpu==2.0.0.alpha0
In [ ]:
# !pip install tqdm
Requirement already satisfied: tqdm in /usr/local/lib/python3.6/dist-packages (4.28.1)
Downloading the Dogs vs Cats dataset¶
In [1]:
!wget --no-check-certificate \
https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip \
-O ./cats_and_dogs_filtered.zip # ./ == 현재경로
--2023-01-02 01:53:51-- https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip Resolving storage.googleapis.com (storage.googleapis.com)... 142.250.99.128, 74.125.142.128, 74.125.195.128, ... Connecting to storage.googleapis.com (storage.googleapis.com)|142.250.99.128|:443... connected. HTTP request sent, awaiting response... 200 OK Length: 68606236 (65M) [application/zip] Saving to: ‘./cats_and_dogs_filtered.zip’ ./cats_and_dogs_fil 100%[===================>] 65.43M 232MB/s in 0.3s 2023-01-02 01:53:51 (232 MB/s) - ‘./cats_and_dogs_filtered.zip’ saved [68606236/68606236]
Stage 2: Dataset preprocessing¶
Import project dependencies¶
In [2]:
import os
import zipfile
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
# 파이썬의 진행상태를 표시해 주는 라이브러리
from tqdm import tqdm_notebook
from tensorflow.keras.preprocessing.image import ImageDataGenerator
%matplotlib inline
tf.__version__
Out[2]:
'2.9.2'
Unzipping the Dogs vs Cats dataset¶
In [4]:
file = zipfile.ZipFile('/content/cats_and_dogs_filtered.zip')
In [5]:
file.extractall('./')
In [ ]:
Seting up dataset paths¶
In [7]:
train_dir = '/content/cats_and_dogs_filtered/train'
test_dir = '/content/cats_and_dogs_filtered/validation'
In [ ]:
In [ ]:
Building the model : MobileNetV2 를 활용¶
Loading the pre-trained model (MobileNetV2)¶
모바일이나, 임베디드에서도 실시간을 작동할 수 있게 모델이 경량화 되면서도, 정확도 또한 많이 떨어지지 않게하여, 속도와 정확도 사이의 트레이드 오프 문제를 어느정도 해결한 네트워크 입니다.
In [8]:
# 우리가 만들려는 모델의 인풋 이미지는 128,128,3 으로 한다.
In [9]:
IMG_SHAPE = (128, 128, 3)
In [ ]:
# 트랜스퍼 러닝은
# 학습이 잘 된 모델을 가져와서, 우리의 문제에 맞게 활용하는 것이므로,
# 학습이 잘 된 모델의 베이스 모델만 가져온다.(헤드모델은 빼고)
In [13]:
base_model = tf.keras.applications.MobileNetV2(input_shape=IMG_SHAPE, include_top=False ) # includ_top=False == 헤드는 빼고 가져와라.
Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/mobilenet_v2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_128_no_top.h5 9406464/9406464 [==============================] - 0s 0us/step
In [14]:
base_model.summary()
Model: "mobilenetv2_1.00_128"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 128, 128, 3 0 []
)]
Conv1 (Conv2D) (None, 64, 64, 32) 864 ['input_1[0][0]']
bn_Conv1 (BatchNormalization) (None, 64, 64, 32) 128 ['Conv1[0][0]']
Conv1_relu (ReLU) (None, 64, 64, 32) 0 ['bn_Conv1[0][0]']
expanded_conv_depthwise (Depth (None, 64, 64, 32) 288 ['Conv1_relu[0][0]']
wiseConv2D)
expanded_conv_depthwise_BN (Ba (None, 64, 64, 32) 128 ['expanded_conv_depthwise[0][0]']
tchNormalization)
expanded_conv_depthwise_relu ( (None, 64, 64, 32) 0 ['expanded_conv_depthwise_BN[0][0
ReLU) ]']
expanded_conv_project (Conv2D) (None, 64, 64, 16) 512 ['expanded_conv_depthwise_relu[0]
[0]']
expanded_conv_project_BN (Batc (None, 64, 64, 16) 64 ['expanded_conv_project[0][0]']
hNormalization)
block_1_expand (Conv2D) (None, 64, 64, 96) 1536 ['expanded_conv_project_BN[0][0]'
]
block_1_expand_BN (BatchNormal (None, 64, 64, 96) 384 ['block_1_expand[0][0]']
ization)
block_1_expand_relu (ReLU) (None, 64, 64, 96) 0 ['block_1_expand_BN[0][0]']
block_1_pad (ZeroPadding2D) (None, 65, 65, 96) 0 ['block_1_expand_relu[0][0]']
block_1_depthwise (DepthwiseCo (None, 32, 32, 96) 864 ['block_1_pad[0][0]']
nv2D)
block_1_depthwise_BN (BatchNor (None, 32, 32, 96) 384 ['block_1_depthwise[0][0]']
malization)
block_1_depthwise_relu (ReLU) (None, 32, 32, 96) 0 ['block_1_depthwise_BN[0][0]']
block_1_project (Conv2D) (None, 32, 32, 24) 2304 ['block_1_depthwise_relu[0][0]']
block_1_project_BN (BatchNorma (None, 32, 32, 24) 96 ['block_1_project[0][0]']
lization)
block_2_expand (Conv2D) (None, 32, 32, 144) 3456 ['block_1_project_BN[0][0]']
block_2_expand_BN (BatchNormal (None, 32, 32, 144) 576 ['block_2_expand[0][0]']
ization)
block_2_expand_relu (ReLU) (None, 32, 32, 144) 0 ['block_2_expand_BN[0][0]']
block_2_depthwise (DepthwiseCo (None, 32, 32, 144) 1296 ['block_2_expand_relu[0][0]']
nv2D)
block_2_depthwise_BN (BatchNor (None, 32, 32, 144) 576 ['block_2_depthwise[0][0]']
malization)
block_2_depthwise_relu (ReLU) (None, 32, 32, 144) 0 ['block_2_depthwise_BN[0][0]']
block_2_project (Conv2D) (None, 32, 32, 24) 3456 ['block_2_depthwise_relu[0][0]']
block_2_project_BN (BatchNorma (None, 32, 32, 24) 96 ['block_2_project[0][0]']
lization)
block_2_add (Add) (None, 32, 32, 24) 0 ['block_1_project_BN[0][0]',
'block_2_project_BN[0][0]']
block_3_expand (Conv2D) (None, 32, 32, 144) 3456 ['block_2_add[0][0]']
block_3_expand_BN (BatchNormal (None, 32, 32, 144) 576 ['block_3_expand[0][0]']
ization)
block_3_expand_relu (ReLU) (None, 32, 32, 144) 0 ['block_3_expand_BN[0][0]']
block_3_pad (ZeroPadding2D) (None, 33, 33, 144) 0 ['block_3_expand_relu[0][0]']
block_3_depthwise (DepthwiseCo (None, 16, 16, 144) 1296 ['block_3_pad[0][0]']
nv2D)
block_3_depthwise_BN (BatchNor (None, 16, 16, 144) 576 ['block_3_depthwise[0][0]']
malization)
block_3_depthwise_relu (ReLU) (None, 16, 16, 144) 0 ['block_3_depthwise_BN[0][0]']
block_3_project (Conv2D) (None, 16, 16, 32) 4608 ['block_3_depthwise_relu[0][0]']
block_3_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_3_project[0][0]']
lization)
block_4_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_3_project_BN[0][0]']
block_4_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_4_expand[0][0]']
ization)
block_4_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_4_expand_BN[0][0]']
block_4_depthwise (DepthwiseCo (None, 16, 16, 192) 1728 ['block_4_expand_relu[0][0]']
nv2D)
block_4_depthwise_BN (BatchNor (None, 16, 16, 192) 768 ['block_4_depthwise[0][0]']
malization)
block_4_depthwise_relu (ReLU) (None, 16, 16, 192) 0 ['block_4_depthwise_BN[0][0]']
block_4_project (Conv2D) (None, 16, 16, 32) 6144 ['block_4_depthwise_relu[0][0]']
block_4_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_4_project[0][0]']
lization)
block_4_add (Add) (None, 16, 16, 32) 0 ['block_3_project_BN[0][0]',
'block_4_project_BN[0][0]']
block_5_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_4_add[0][0]']
block_5_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_5_expand[0][0]']
ization)
block_5_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_5_expand_BN[0][0]']
block_5_depthwise (DepthwiseCo (None, 16, 16, 192) 1728 ['block_5_expand_relu[0][0]']
nv2D)
block_5_depthwise_BN (BatchNor (None, 16, 16, 192) 768 ['block_5_depthwise[0][0]']
malization)
block_5_depthwise_relu (ReLU) (None, 16, 16, 192) 0 ['block_5_depthwise_BN[0][0]']
block_5_project (Conv2D) (None, 16, 16, 32) 6144 ['block_5_depthwise_relu[0][0]']
block_5_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_5_project[0][0]']
lization)
block_5_add (Add) (None, 16, 16, 32) 0 ['block_4_add[0][0]',
'block_5_project_BN[0][0]']
block_6_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_5_add[0][0]']
block_6_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_6_expand[0][0]']
ization)
block_6_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_6_expand_BN[0][0]']
block_6_pad (ZeroPadding2D) (None, 17, 17, 192) 0 ['block_6_expand_relu[0][0]']
block_6_depthwise (DepthwiseCo (None, 8, 8, 192) 1728 ['block_6_pad[0][0]']
nv2D)
block_6_depthwise_BN (BatchNor (None, 8, 8, 192) 768 ['block_6_depthwise[0][0]']
malization)
block_6_depthwise_relu (ReLU) (None, 8, 8, 192) 0 ['block_6_depthwise_BN[0][0]']
block_6_project (Conv2D) (None, 8, 8, 64) 12288 ['block_6_depthwise_relu[0][0]']
block_6_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_6_project[0][0]']
lization)
block_7_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_6_project_BN[0][0]']
block_7_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_7_expand[0][0]']
ization)
block_7_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_7_expand_BN[0][0]']
block_7_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_7_expand_relu[0][0]']
nv2D)
block_7_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_7_depthwise[0][0]']
malization)
block_7_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_7_depthwise_BN[0][0]']
block_7_project (Conv2D) (None, 8, 8, 64) 24576 ['block_7_depthwise_relu[0][0]']
block_7_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_7_project[0][0]']
lization)
block_7_add (Add) (None, 8, 8, 64) 0 ['block_6_project_BN[0][0]',
'block_7_project_BN[0][0]']
block_8_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_7_add[0][0]']
block_8_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_8_expand[0][0]']
ization)
block_8_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_8_expand_BN[0][0]']
block_8_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_8_expand_relu[0][0]']
nv2D)
block_8_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_8_depthwise[0][0]']
malization)
block_8_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_8_depthwise_BN[0][0]']
block_8_project (Conv2D) (None, 8, 8, 64) 24576 ['block_8_depthwise_relu[0][0]']
block_8_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_8_project[0][0]']
lization)
block_8_add (Add) (None, 8, 8, 64) 0 ['block_7_add[0][0]',
'block_8_project_BN[0][0]']
block_9_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_8_add[0][0]']
block_9_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_9_expand[0][0]']
ization)
block_9_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_9_expand_BN[0][0]']
block_9_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_9_expand_relu[0][0]']
nv2D)
block_9_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_9_depthwise[0][0]']
malization)
block_9_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_9_depthwise_BN[0][0]']
block_9_project (Conv2D) (None, 8, 8, 64) 24576 ['block_9_depthwise_relu[0][0]']
block_9_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_9_project[0][0]']
lization)
block_9_add (Add) (None, 8, 8, 64) 0 ['block_8_add[0][0]',
'block_9_project_BN[0][0]']
block_10_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_9_add[0][0]']
block_10_expand_BN (BatchNorma (None, 8, 8, 384) 1536 ['block_10_expand[0][0]']
lization)
block_10_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_10_expand_BN[0][0]']
block_10_depthwise (DepthwiseC (None, 8, 8, 384) 3456 ['block_10_expand_relu[0][0]']
onv2D)
block_10_depthwise_BN (BatchNo (None, 8, 8, 384) 1536 ['block_10_depthwise[0][0]']
rmalization)
block_10_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_10_depthwise_BN[0][0]']
block_10_project (Conv2D) (None, 8, 8, 96) 36864 ['block_10_depthwise_relu[0][0]']
block_10_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_10_project[0][0]']
alization)
block_11_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_10_project_BN[0][0]']
block_11_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_11_expand[0][0]']
lization)
block_11_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_11_expand_BN[0][0]']
block_11_depthwise (DepthwiseC (None, 8, 8, 576) 5184 ['block_11_expand_relu[0][0]']
onv2D)
block_11_depthwise_BN (BatchNo (None, 8, 8, 576) 2304 ['block_11_depthwise[0][0]']
rmalization)
block_11_depthwise_relu (ReLU) (None, 8, 8, 576) 0 ['block_11_depthwise_BN[0][0]']
block_11_project (Conv2D) (None, 8, 8, 96) 55296 ['block_11_depthwise_relu[0][0]']
block_11_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_11_project[0][0]']
alization)
block_11_add (Add) (None, 8, 8, 96) 0 ['block_10_project_BN[0][0]',
'block_11_project_BN[0][0]']
block_12_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_11_add[0][0]']
block_12_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_12_expand[0][0]']
lization)
block_12_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_12_expand_BN[0][0]']
block_12_depthwise (DepthwiseC (None, 8, 8, 576) 5184 ['block_12_expand_relu[0][0]']
onv2D)
block_12_depthwise_BN (BatchNo (None, 8, 8, 576) 2304 ['block_12_depthwise[0][0]']
rmalization)
block_12_depthwise_relu (ReLU) (None, 8, 8, 576) 0 ['block_12_depthwise_BN[0][0]']
block_12_project (Conv2D) (None, 8, 8, 96) 55296 ['block_12_depthwise_relu[0][0]']
block_12_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_12_project[0][0]']
alization)
block_12_add (Add) (None, 8, 8, 96) 0 ['block_11_add[0][0]',
'block_12_project_BN[0][0]']
block_13_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_12_add[0][0]']
block_13_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_13_expand[0][0]']
lization)
block_13_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_13_expand_BN[0][0]']
block_13_pad (ZeroPadding2D) (None, 9, 9, 576) 0 ['block_13_expand_relu[0][0]']
block_13_depthwise (DepthwiseC (None, 4, 4, 576) 5184 ['block_13_pad[0][0]']
onv2D)
block_13_depthwise_BN (BatchNo (None, 4, 4, 576) 2304 ['block_13_depthwise[0][0]']
rmalization)
block_13_depthwise_relu (ReLU) (None, 4, 4, 576) 0 ['block_13_depthwise_BN[0][0]']
block_13_project (Conv2D) (None, 4, 4, 160) 92160 ['block_13_depthwise_relu[0][0]']
block_13_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_13_project[0][0]']
alization)
block_14_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_13_project_BN[0][0]']
block_14_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_14_expand[0][0]']
lization)
block_14_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_14_expand_BN[0][0]']
block_14_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_14_expand_relu[0][0]']
onv2D)
block_14_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_14_depthwise[0][0]']
rmalization)
block_14_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_14_depthwise_BN[0][0]']
block_14_project (Conv2D) (None, 4, 4, 160) 153600 ['block_14_depthwise_relu[0][0]']
block_14_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_14_project[0][0]']
alization)
block_14_add (Add) (None, 4, 4, 160) 0 ['block_13_project_BN[0][0]',
'block_14_project_BN[0][0]']
block_15_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_14_add[0][0]']
block_15_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_15_expand[0][0]']
lization)
block_15_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_15_expand_BN[0][0]']
block_15_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_15_expand_relu[0][0]']
onv2D)
block_15_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_15_depthwise[0][0]']
rmalization)
block_15_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_15_depthwise_BN[0][0]']
block_15_project (Conv2D) (None, 4, 4, 160) 153600 ['block_15_depthwise_relu[0][0]']
block_15_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_15_project[0][0]']
alization)
block_15_add (Add) (None, 4, 4, 160) 0 ['block_14_add[0][0]',
'block_15_project_BN[0][0]']
block_16_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_15_add[0][0]']
block_16_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_16_expand[0][0]']
lization)
block_16_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_16_expand_BN[0][0]']
block_16_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_16_expand_relu[0][0]']
onv2D)
block_16_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_16_depthwise[0][0]']
rmalization)
block_16_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_16_depthwise_BN[0][0]']
block_16_project (Conv2D) (None, 4, 4, 320) 307200 ['block_16_depthwise_relu[0][0]']
block_16_project_BN (BatchNorm (None, 4, 4, 320) 1280 ['block_16_project[0][0]']
alization)
Conv_1 (Conv2D) (None, 4, 4, 1280) 409600 ['block_16_project_BN[0][0]']
Conv_1_bn (BatchNormalization) (None, 4, 4, 1280) 5120 ['Conv_1[0][0]']
out_relu (ReLU) (None, 4, 4, 1280) 0 ['Conv_1_bn[0][0]']
==================================================================================================
Total params: 2,257,984
Trainable params: 2,223,872
Non-trainable params: 34,112
__________________________________________________________________________________________________
In [ ]:
In [ ]:
Freezing the base model¶
In [ ]:
# 우리가 가져온 모델의, 베이스 모델 부분은
# 이미 1400만장 이상의 이미지로 학습이 잘 되어서,
# 이미지의 특징을 뽑아내는 역할을 하므로,
# 우리 데이터로는 학습이 되지 않도록 한다.
In [15]:
base_model.summary()
Model: "mobilenetv2_1.00_128"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 128, 128, 3 0 []
)]
Conv1 (Conv2D) (None, 64, 64, 32) 864 ['input_1[0][0]']
bn_Conv1 (BatchNormalization) (None, 64, 64, 32) 128 ['Conv1[0][0]']
Conv1_relu (ReLU) (None, 64, 64, 32) 0 ['bn_Conv1[0][0]']
expanded_conv_depthwise (Depth (None, 64, 64, 32) 288 ['Conv1_relu[0][0]']
wiseConv2D)
expanded_conv_depthwise_BN (Ba (None, 64, 64, 32) 128 ['expanded_conv_depthwise[0][0]']
tchNormalization)
expanded_conv_depthwise_relu ( (None, 64, 64, 32) 0 ['expanded_conv_depthwise_BN[0][0
ReLU) ]']
expanded_conv_project (Conv2D) (None, 64, 64, 16) 512 ['expanded_conv_depthwise_relu[0]
[0]']
expanded_conv_project_BN (Batc (None, 64, 64, 16) 64 ['expanded_conv_project[0][0]']
hNormalization)
block_1_expand (Conv2D) (None, 64, 64, 96) 1536 ['expanded_conv_project_BN[0][0]'
]
block_1_expand_BN (BatchNormal (None, 64, 64, 96) 384 ['block_1_expand[0][0]']
ization)
block_1_expand_relu (ReLU) (None, 64, 64, 96) 0 ['block_1_expand_BN[0][0]']
block_1_pad (ZeroPadding2D) (None, 65, 65, 96) 0 ['block_1_expand_relu[0][0]']
block_1_depthwise (DepthwiseCo (None, 32, 32, 96) 864 ['block_1_pad[0][0]']
nv2D)
block_1_depthwise_BN (BatchNor (None, 32, 32, 96) 384 ['block_1_depthwise[0][0]']
malization)
block_1_depthwise_relu (ReLU) (None, 32, 32, 96) 0 ['block_1_depthwise_BN[0][0]']
block_1_project (Conv2D) (None, 32, 32, 24) 2304 ['block_1_depthwise_relu[0][0]']
block_1_project_BN (BatchNorma (None, 32, 32, 24) 96 ['block_1_project[0][0]']
lization)
block_2_expand (Conv2D) (None, 32, 32, 144) 3456 ['block_1_project_BN[0][0]']
block_2_expand_BN (BatchNormal (None, 32, 32, 144) 576 ['block_2_expand[0][0]']
ization)
block_2_expand_relu (ReLU) (None, 32, 32, 144) 0 ['block_2_expand_BN[0][0]']
block_2_depthwise (DepthwiseCo (None, 32, 32, 144) 1296 ['block_2_expand_relu[0][0]']
nv2D)
block_2_depthwise_BN (BatchNor (None, 32, 32, 144) 576 ['block_2_depthwise[0][0]']
malization)
block_2_depthwise_relu (ReLU) (None, 32, 32, 144) 0 ['block_2_depthwise_BN[0][0]']
block_2_project (Conv2D) (None, 32, 32, 24) 3456 ['block_2_depthwise_relu[0][0]']
block_2_project_BN (BatchNorma (None, 32, 32, 24) 96 ['block_2_project[0][0]']
lization)
block_2_add (Add) (None, 32, 32, 24) 0 ['block_1_project_BN[0][0]',
'block_2_project_BN[0][0]']
block_3_expand (Conv2D) (None, 32, 32, 144) 3456 ['block_2_add[0][0]']
block_3_expand_BN (BatchNormal (None, 32, 32, 144) 576 ['block_3_expand[0][0]']
ization)
block_3_expand_relu (ReLU) (None, 32, 32, 144) 0 ['block_3_expand_BN[0][0]']
block_3_pad (ZeroPadding2D) (None, 33, 33, 144) 0 ['block_3_expand_relu[0][0]']
block_3_depthwise (DepthwiseCo (None, 16, 16, 144) 1296 ['block_3_pad[0][0]']
nv2D)
block_3_depthwise_BN (BatchNor (None, 16, 16, 144) 576 ['block_3_depthwise[0][0]']
malization)
block_3_depthwise_relu (ReLU) (None, 16, 16, 144) 0 ['block_3_depthwise_BN[0][0]']
block_3_project (Conv2D) (None, 16, 16, 32) 4608 ['block_3_depthwise_relu[0][0]']
block_3_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_3_project[0][0]']
lization)
block_4_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_3_project_BN[0][0]']
block_4_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_4_expand[0][0]']
ization)
block_4_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_4_expand_BN[0][0]']
block_4_depthwise (DepthwiseCo (None, 16, 16, 192) 1728 ['block_4_expand_relu[0][0]']
nv2D)
block_4_depthwise_BN (BatchNor (None, 16, 16, 192) 768 ['block_4_depthwise[0][0]']
malization)
block_4_depthwise_relu (ReLU) (None, 16, 16, 192) 0 ['block_4_depthwise_BN[0][0]']
block_4_project (Conv2D) (None, 16, 16, 32) 6144 ['block_4_depthwise_relu[0][0]']
block_4_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_4_project[0][0]']
lization)
block_4_add (Add) (None, 16, 16, 32) 0 ['block_3_project_BN[0][0]',
'block_4_project_BN[0][0]']
block_5_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_4_add[0][0]']
block_5_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_5_expand[0][0]']
ization)
block_5_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_5_expand_BN[0][0]']
block_5_depthwise (DepthwiseCo (None, 16, 16, 192) 1728 ['block_5_expand_relu[0][0]']
nv2D)
block_5_depthwise_BN (BatchNor (None, 16, 16, 192) 768 ['block_5_depthwise[0][0]']
malization)
block_5_depthwise_relu (ReLU) (None, 16, 16, 192) 0 ['block_5_depthwise_BN[0][0]']
block_5_project (Conv2D) (None, 16, 16, 32) 6144 ['block_5_depthwise_relu[0][0]']
block_5_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_5_project[0][0]']
lization)
block_5_add (Add) (None, 16, 16, 32) 0 ['block_4_add[0][0]',
'block_5_project_BN[0][0]']
block_6_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_5_add[0][0]']
block_6_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_6_expand[0][0]']
ization)
block_6_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_6_expand_BN[0][0]']
block_6_pad (ZeroPadding2D) (None, 17, 17, 192) 0 ['block_6_expand_relu[0][0]']
block_6_depthwise (DepthwiseCo (None, 8, 8, 192) 1728 ['block_6_pad[0][0]']
nv2D)
block_6_depthwise_BN (BatchNor (None, 8, 8, 192) 768 ['block_6_depthwise[0][0]']
malization)
block_6_depthwise_relu (ReLU) (None, 8, 8, 192) 0 ['block_6_depthwise_BN[0][0]']
block_6_project (Conv2D) (None, 8, 8, 64) 12288 ['block_6_depthwise_relu[0][0]']
block_6_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_6_project[0][0]']
lization)
block_7_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_6_project_BN[0][0]']
block_7_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_7_expand[0][0]']
ization)
block_7_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_7_expand_BN[0][0]']
block_7_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_7_expand_relu[0][0]']
nv2D)
block_7_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_7_depthwise[0][0]']
malization)
block_7_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_7_depthwise_BN[0][0]']
block_7_project (Conv2D) (None, 8, 8, 64) 24576 ['block_7_depthwise_relu[0][0]']
block_7_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_7_project[0][0]']
lization)
block_7_add (Add) (None, 8, 8, 64) 0 ['block_6_project_BN[0][0]',
'block_7_project_BN[0][0]']
block_8_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_7_add[0][0]']
block_8_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_8_expand[0][0]']
ization)
block_8_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_8_expand_BN[0][0]']
block_8_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_8_expand_relu[0][0]']
nv2D)
block_8_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_8_depthwise[0][0]']
malization)
block_8_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_8_depthwise_BN[0][0]']
block_8_project (Conv2D) (None, 8, 8, 64) 24576 ['block_8_depthwise_relu[0][0]']
block_8_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_8_project[0][0]']
lization)
block_8_add (Add) (None, 8, 8, 64) 0 ['block_7_add[0][0]',
'block_8_project_BN[0][0]']
block_9_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_8_add[0][0]']
block_9_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_9_expand[0][0]']
ization)
block_9_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_9_expand_BN[0][0]']
block_9_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_9_expand_relu[0][0]']
nv2D)
block_9_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_9_depthwise[0][0]']
malization)
block_9_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_9_depthwise_BN[0][0]']
block_9_project (Conv2D) (None, 8, 8, 64) 24576 ['block_9_depthwise_relu[0][0]']
block_9_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_9_project[0][0]']
lization)
block_9_add (Add) (None, 8, 8, 64) 0 ['block_8_add[0][0]',
'block_9_project_BN[0][0]']
block_10_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_9_add[0][0]']
block_10_expand_BN (BatchNorma (None, 8, 8, 384) 1536 ['block_10_expand[0][0]']
lization)
block_10_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_10_expand_BN[0][0]']
block_10_depthwise (DepthwiseC (None, 8, 8, 384) 3456 ['block_10_expand_relu[0][0]']
onv2D)
block_10_depthwise_BN (BatchNo (None, 8, 8, 384) 1536 ['block_10_depthwise[0][0]']
rmalization)
block_10_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_10_depthwise_BN[0][0]']
block_10_project (Conv2D) (None, 8, 8, 96) 36864 ['block_10_depthwise_relu[0][0]']
block_10_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_10_project[0][0]']
alization)
block_11_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_10_project_BN[0][0]']
block_11_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_11_expand[0][0]']
lization)
block_11_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_11_expand_BN[0][0]']
block_11_depthwise (DepthwiseC (None, 8, 8, 576) 5184 ['block_11_expand_relu[0][0]']
onv2D)
block_11_depthwise_BN (BatchNo (None, 8, 8, 576) 2304 ['block_11_depthwise[0][0]']
rmalization)
block_11_depthwise_relu (ReLU) (None, 8, 8, 576) 0 ['block_11_depthwise_BN[0][0]']
block_11_project (Conv2D) (None, 8, 8, 96) 55296 ['block_11_depthwise_relu[0][0]']
block_11_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_11_project[0][0]']
alization)
block_11_add (Add) (None, 8, 8, 96) 0 ['block_10_project_BN[0][0]',
'block_11_project_BN[0][0]']
block_12_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_11_add[0][0]']
block_12_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_12_expand[0][0]']
lization)
block_12_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_12_expand_BN[0][0]']
block_12_depthwise (DepthwiseC (None, 8, 8, 576) 5184 ['block_12_expand_relu[0][0]']
onv2D)
block_12_depthwise_BN (BatchNo (None, 8, 8, 576) 2304 ['block_12_depthwise[0][0]']
rmalization)
block_12_depthwise_relu (ReLU) (None, 8, 8, 576) 0 ['block_12_depthwise_BN[0][0]']
block_12_project (Conv2D) (None, 8, 8, 96) 55296 ['block_12_depthwise_relu[0][0]']
block_12_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_12_project[0][0]']
alization)
block_12_add (Add) (None, 8, 8, 96) 0 ['block_11_add[0][0]',
'block_12_project_BN[0][0]']
block_13_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_12_add[0][0]']
block_13_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_13_expand[0][0]']
lization)
block_13_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_13_expand_BN[0][0]']
block_13_pad (ZeroPadding2D) (None, 9, 9, 576) 0 ['block_13_expand_relu[0][0]']
block_13_depthwise (DepthwiseC (None, 4, 4, 576) 5184 ['block_13_pad[0][0]']
onv2D)
block_13_depthwise_BN (BatchNo (None, 4, 4, 576) 2304 ['block_13_depthwise[0][0]']
rmalization)
block_13_depthwise_relu (ReLU) (None, 4, 4, 576) 0 ['block_13_depthwise_BN[0][0]']
block_13_project (Conv2D) (None, 4, 4, 160) 92160 ['block_13_depthwise_relu[0][0]']
block_13_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_13_project[0][0]']
alization)
block_14_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_13_project_BN[0][0]']
block_14_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_14_expand[0][0]']
lization)
block_14_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_14_expand_BN[0][0]']
block_14_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_14_expand_relu[0][0]']
onv2D)
block_14_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_14_depthwise[0][0]']
rmalization)
block_14_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_14_depthwise_BN[0][0]']
block_14_project (Conv2D) (None, 4, 4, 160) 153600 ['block_14_depthwise_relu[0][0]']
block_14_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_14_project[0][0]']
alization)
block_14_add (Add) (None, 4, 4, 160) 0 ['block_13_project_BN[0][0]',
'block_14_project_BN[0][0]']
block_15_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_14_add[0][0]']
block_15_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_15_expand[0][0]']
lization)
block_15_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_15_expand_BN[0][0]']
block_15_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_15_expand_relu[0][0]']
onv2D)
block_15_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_15_depthwise[0][0]']
rmalization)
block_15_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_15_depthwise_BN[0][0]']
block_15_project (Conv2D) (None, 4, 4, 160) 153600 ['block_15_depthwise_relu[0][0]']
block_15_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_15_project[0][0]']
alization)
block_15_add (Add) (None, 4, 4, 160) 0 ['block_14_add[0][0]',
'block_15_project_BN[0][0]']
block_16_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_15_add[0][0]']
block_16_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_16_expand[0][0]']
lization)
block_16_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_16_expand_BN[0][0]']
block_16_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_16_expand_relu[0][0]']
onv2D)
block_16_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_16_depthwise[0][0]']
rmalization)
block_16_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_16_depthwise_BN[0][0]']
block_16_project (Conv2D) (None, 4, 4, 320) 307200 ['block_16_depthwise_relu[0][0]']
block_16_project_BN (BatchNorm (None, 4, 4, 320) 1280 ['block_16_project[0][0]']
alization)
Conv_1 (Conv2D) (None, 4, 4, 1280) 409600 ['block_16_project_BN[0][0]']
Conv_1_bn (BatchNormalization) (None, 4, 4, 1280) 5120 ['Conv_1[0][0]']
out_relu (ReLU) (None, 4, 4, 1280) 0 ['Conv_1_bn[0][0]']
==================================================================================================
Total params: 2,257,984
Trainable params: 2,223,872
Non-trainable params: 34,112
__________________________________________________________________________________________________
In [16]:
base_model.trainable= False
In [17]:
base_model.summary()
Model: "mobilenetv2_1.00_128"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 128, 128, 3 0 []
)]
Conv1 (Conv2D) (None, 64, 64, 32) 864 ['input_1[0][0]']
bn_Conv1 (BatchNormalization) (None, 64, 64, 32) 128 ['Conv1[0][0]']
Conv1_relu (ReLU) (None, 64, 64, 32) 0 ['bn_Conv1[0][0]']
expanded_conv_depthwise (Depth (None, 64, 64, 32) 288 ['Conv1_relu[0][0]']
wiseConv2D)
expanded_conv_depthwise_BN (Ba (None, 64, 64, 32) 128 ['expanded_conv_depthwise[0][0]']
tchNormalization)
expanded_conv_depthwise_relu ( (None, 64, 64, 32) 0 ['expanded_conv_depthwise_BN[0][0
ReLU) ]']
expanded_conv_project (Conv2D) (None, 64, 64, 16) 512 ['expanded_conv_depthwise_relu[0]
[0]']
expanded_conv_project_BN (Batc (None, 64, 64, 16) 64 ['expanded_conv_project[0][0]']
hNormalization)
block_1_expand (Conv2D) (None, 64, 64, 96) 1536 ['expanded_conv_project_BN[0][0]'
]
block_1_expand_BN (BatchNormal (None, 64, 64, 96) 384 ['block_1_expand[0][0]']
ization)
block_1_expand_relu (ReLU) (None, 64, 64, 96) 0 ['block_1_expand_BN[0][0]']
block_1_pad (ZeroPadding2D) (None, 65, 65, 96) 0 ['block_1_expand_relu[0][0]']
block_1_depthwise (DepthwiseCo (None, 32, 32, 96) 864 ['block_1_pad[0][0]']
nv2D)
block_1_depthwise_BN (BatchNor (None, 32, 32, 96) 384 ['block_1_depthwise[0][0]']
malization)
block_1_depthwise_relu (ReLU) (None, 32, 32, 96) 0 ['block_1_depthwise_BN[0][0]']
block_1_project (Conv2D) (None, 32, 32, 24) 2304 ['block_1_depthwise_relu[0][0]']
block_1_project_BN (BatchNorma (None, 32, 32, 24) 96 ['block_1_project[0][0]']
lization)
block_2_expand (Conv2D) (None, 32, 32, 144) 3456 ['block_1_project_BN[0][0]']
block_2_expand_BN (BatchNormal (None, 32, 32, 144) 576 ['block_2_expand[0][0]']
ization)
block_2_expand_relu (ReLU) (None, 32, 32, 144) 0 ['block_2_expand_BN[0][0]']
block_2_depthwise (DepthwiseCo (None, 32, 32, 144) 1296 ['block_2_expand_relu[0][0]']
nv2D)
block_2_depthwise_BN (BatchNor (None, 32, 32, 144) 576 ['block_2_depthwise[0][0]']
malization)
block_2_depthwise_relu (ReLU) (None, 32, 32, 144) 0 ['block_2_depthwise_BN[0][0]']
block_2_project (Conv2D) (None, 32, 32, 24) 3456 ['block_2_depthwise_relu[0][0]']
block_2_project_BN (BatchNorma (None, 32, 32, 24) 96 ['block_2_project[0][0]']
lization)
block_2_add (Add) (None, 32, 32, 24) 0 ['block_1_project_BN[0][0]',
'block_2_project_BN[0][0]']
block_3_expand (Conv2D) (None, 32, 32, 144) 3456 ['block_2_add[0][0]']
block_3_expand_BN (BatchNormal (None, 32, 32, 144) 576 ['block_3_expand[0][0]']
ization)
block_3_expand_relu (ReLU) (None, 32, 32, 144) 0 ['block_3_expand_BN[0][0]']
block_3_pad (ZeroPadding2D) (None, 33, 33, 144) 0 ['block_3_expand_relu[0][0]']
block_3_depthwise (DepthwiseCo (None, 16, 16, 144) 1296 ['block_3_pad[0][0]']
nv2D)
block_3_depthwise_BN (BatchNor (None, 16, 16, 144) 576 ['block_3_depthwise[0][0]']
malization)
block_3_depthwise_relu (ReLU) (None, 16, 16, 144) 0 ['block_3_depthwise_BN[0][0]']
block_3_project (Conv2D) (None, 16, 16, 32) 4608 ['block_3_depthwise_relu[0][0]']
block_3_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_3_project[0][0]']
lization)
block_4_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_3_project_BN[0][0]']
block_4_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_4_expand[0][0]']
ization)
block_4_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_4_expand_BN[0][0]']
block_4_depthwise (DepthwiseCo (None, 16, 16, 192) 1728 ['block_4_expand_relu[0][0]']
nv2D)
block_4_depthwise_BN (BatchNor (None, 16, 16, 192) 768 ['block_4_depthwise[0][0]']
malization)
block_4_depthwise_relu (ReLU) (None, 16, 16, 192) 0 ['block_4_depthwise_BN[0][0]']
block_4_project (Conv2D) (None, 16, 16, 32) 6144 ['block_4_depthwise_relu[0][0]']
block_4_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_4_project[0][0]']
lization)
block_4_add (Add) (None, 16, 16, 32) 0 ['block_3_project_BN[0][0]',
'block_4_project_BN[0][0]']
block_5_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_4_add[0][0]']
block_5_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_5_expand[0][0]']
ization)
block_5_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_5_expand_BN[0][0]']
block_5_depthwise (DepthwiseCo (None, 16, 16, 192) 1728 ['block_5_expand_relu[0][0]']
nv2D)
block_5_depthwise_BN (BatchNor (None, 16, 16, 192) 768 ['block_5_depthwise[0][0]']
malization)
block_5_depthwise_relu (ReLU) (None, 16, 16, 192) 0 ['block_5_depthwise_BN[0][0]']
block_5_project (Conv2D) (None, 16, 16, 32) 6144 ['block_5_depthwise_relu[0][0]']
block_5_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_5_project[0][0]']
lization)
block_5_add (Add) (None, 16, 16, 32) 0 ['block_4_add[0][0]',
'block_5_project_BN[0][0]']
block_6_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_5_add[0][0]']
block_6_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_6_expand[0][0]']
ization)
block_6_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_6_expand_BN[0][0]']
block_6_pad (ZeroPadding2D) (None, 17, 17, 192) 0 ['block_6_expand_relu[0][0]']
block_6_depthwise (DepthwiseCo (None, 8, 8, 192) 1728 ['block_6_pad[0][0]']
nv2D)
block_6_depthwise_BN (BatchNor (None, 8, 8, 192) 768 ['block_6_depthwise[0][0]']
malization)
block_6_depthwise_relu (ReLU) (None, 8, 8, 192) 0 ['block_6_depthwise_BN[0][0]']
block_6_project (Conv2D) (None, 8, 8, 64) 12288 ['block_6_depthwise_relu[0][0]']
block_6_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_6_project[0][0]']
lization)
block_7_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_6_project_BN[0][0]']
block_7_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_7_expand[0][0]']
ization)
block_7_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_7_expand_BN[0][0]']
block_7_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_7_expand_relu[0][0]']
nv2D)
block_7_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_7_depthwise[0][0]']
malization)
block_7_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_7_depthwise_BN[0][0]']
block_7_project (Conv2D) (None, 8, 8, 64) 24576 ['block_7_depthwise_relu[0][0]']
block_7_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_7_project[0][0]']
lization)
block_7_add (Add) (None, 8, 8, 64) 0 ['block_6_project_BN[0][0]',
'block_7_project_BN[0][0]']
block_8_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_7_add[0][0]']
block_8_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_8_expand[0][0]']
ization)
block_8_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_8_expand_BN[0][0]']
block_8_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_8_expand_relu[0][0]']
nv2D)
block_8_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_8_depthwise[0][0]']
malization)
block_8_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_8_depthwise_BN[0][0]']
block_8_project (Conv2D) (None, 8, 8, 64) 24576 ['block_8_depthwise_relu[0][0]']
block_8_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_8_project[0][0]']
lization)
block_8_add (Add) (None, 8, 8, 64) 0 ['block_7_add[0][0]',
'block_8_project_BN[0][0]']
block_9_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_8_add[0][0]']
block_9_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_9_expand[0][0]']
ization)
block_9_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_9_expand_BN[0][0]']
block_9_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_9_expand_relu[0][0]']
nv2D)
block_9_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_9_depthwise[0][0]']
malization)
block_9_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_9_depthwise_BN[0][0]']
block_9_project (Conv2D) (None, 8, 8, 64) 24576 ['block_9_depthwise_relu[0][0]']
block_9_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_9_project[0][0]']
lization)
block_9_add (Add) (None, 8, 8, 64) 0 ['block_8_add[0][0]',
'block_9_project_BN[0][0]']
block_10_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_9_add[0][0]']
block_10_expand_BN (BatchNorma (None, 8, 8, 384) 1536 ['block_10_expand[0][0]']
lization)
block_10_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_10_expand_BN[0][0]']
block_10_depthwise (DepthwiseC (None, 8, 8, 384) 3456 ['block_10_expand_relu[0][0]']
onv2D)
block_10_depthwise_BN (BatchNo (None, 8, 8, 384) 1536 ['block_10_depthwise[0][0]']
rmalization)
block_10_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_10_depthwise_BN[0][0]']
block_10_project (Conv2D) (None, 8, 8, 96) 36864 ['block_10_depthwise_relu[0][0]']
block_10_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_10_project[0][0]']
alization)
block_11_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_10_project_BN[0][0]']
block_11_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_11_expand[0][0]']
lization)
block_11_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_11_expand_BN[0][0]']
block_11_depthwise (DepthwiseC (None, 8, 8, 576) 5184 ['block_11_expand_relu[0][0]']
onv2D)
block_11_depthwise_BN (BatchNo (None, 8, 8, 576) 2304 ['block_11_depthwise[0][0]']
rmalization)
block_11_depthwise_relu (ReLU) (None, 8, 8, 576) 0 ['block_11_depthwise_BN[0][0]']
block_11_project (Conv2D) (None, 8, 8, 96) 55296 ['block_11_depthwise_relu[0][0]']
block_11_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_11_project[0][0]']
alization)
block_11_add (Add) (None, 8, 8, 96) 0 ['block_10_project_BN[0][0]',
'block_11_project_BN[0][0]']
block_12_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_11_add[0][0]']
block_12_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_12_expand[0][0]']
lization)
block_12_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_12_expand_BN[0][0]']
block_12_depthwise (DepthwiseC (None, 8, 8, 576) 5184 ['block_12_expand_relu[0][0]']
onv2D)
block_12_depthwise_BN (BatchNo (None, 8, 8, 576) 2304 ['block_12_depthwise[0][0]']
rmalization)
block_12_depthwise_relu (ReLU) (None, 8, 8, 576) 0 ['block_12_depthwise_BN[0][0]']
block_12_project (Conv2D) (None, 8, 8, 96) 55296 ['block_12_depthwise_relu[0][0]']
block_12_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_12_project[0][0]']
alization)
block_12_add (Add) (None, 8, 8, 96) 0 ['block_11_add[0][0]',
'block_12_project_BN[0][0]']
block_13_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_12_add[0][0]']
block_13_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_13_expand[0][0]']
lization)
block_13_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_13_expand_BN[0][0]']
block_13_pad (ZeroPadding2D) (None, 9, 9, 576) 0 ['block_13_expand_relu[0][0]']
block_13_depthwise (DepthwiseC (None, 4, 4, 576) 5184 ['block_13_pad[0][0]']
onv2D)
block_13_depthwise_BN (BatchNo (None, 4, 4, 576) 2304 ['block_13_depthwise[0][0]']
rmalization)
block_13_depthwise_relu (ReLU) (None, 4, 4, 576) 0 ['block_13_depthwise_BN[0][0]']
block_13_project (Conv2D) (None, 4, 4, 160) 92160 ['block_13_depthwise_relu[0][0]']
block_13_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_13_project[0][0]']
alization)
block_14_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_13_project_BN[0][0]']
block_14_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_14_expand[0][0]']
lization)
block_14_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_14_expand_BN[0][0]']
block_14_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_14_expand_relu[0][0]']
onv2D)
block_14_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_14_depthwise[0][0]']
rmalization)
block_14_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_14_depthwise_BN[0][0]']
block_14_project (Conv2D) (None, 4, 4, 160) 153600 ['block_14_depthwise_relu[0][0]']
block_14_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_14_project[0][0]']
alization)
block_14_add (Add) (None, 4, 4, 160) 0 ['block_13_project_BN[0][0]',
'block_14_project_BN[0][0]']
block_15_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_14_add[0][0]']
block_15_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_15_expand[0][0]']
lization)
block_15_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_15_expand_BN[0][0]']
block_15_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_15_expand_relu[0][0]']
onv2D)
block_15_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_15_depthwise[0][0]']
rmalization)
block_15_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_15_depthwise_BN[0][0]']
block_15_project (Conv2D) (None, 4, 4, 160) 153600 ['block_15_depthwise_relu[0][0]']
block_15_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_15_project[0][0]']
alization)
block_15_add (Add) (None, 4, 4, 160) 0 ['block_14_add[0][0]',
'block_15_project_BN[0][0]']
block_16_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_15_add[0][0]']
block_16_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_16_expand[0][0]']
lization)
block_16_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_16_expand_BN[0][0]']
block_16_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_16_expand_relu[0][0]']
onv2D)
block_16_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_16_depthwise[0][0]']
rmalization)
block_16_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_16_depthwise_BN[0][0]']
block_16_project (Conv2D) (None, 4, 4, 320) 307200 ['block_16_depthwise_relu[0][0]']
block_16_project_BN (BatchNorm (None, 4, 4, 320) 1280 ['block_16_project[0][0]']
alization)
Conv_1 (Conv2D) (None, 4, 4, 1280) 409600 ['block_16_project_BN[0][0]']
Conv_1_bn (BatchNormalization) (None, 4, 4, 1280) 5120 ['Conv_1[0][0]']
out_relu (ReLU) (None, 4, 4, 1280) 0 ['Conv_1_bn[0][0]']
==================================================================================================
Total params: 2,257,984
Trainable params: 0
Non-trainable params: 2,257,984
__________________________________________________________________________________________________
In [ ]:
# Trainable params: 0 == 학습할수있는 파라미터가 0 개가 되었음을 확인할수잇다.
Defining the custom head for our network¶
In [21]:
from keras.layers import Flatten, Dense
In [22]:
head_model = base_model.output
In [23]:
head_model = Flatten()(head_model) # 함수형선언 // 헤드모델뒤에 플래톤을 붙혀라는뜻
In [24]:
head_model = Dense(128,'relu')(head_model)
In [25]:
head_model = Dense(1,'sigmoid')(head_model) # 아웃풋레이어
In [ ]:
Defining the model¶
In [26]:
from keras.models import Model
In [28]:
model = Model( inputs = base_model.input, outputs = head_model) # 베이스모델과 헤드모델을 합치는것
In [29]:
model.summary()
Model: "model_1"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 128, 128, 3 0 []
)]
Conv1 (Conv2D) (None, 64, 64, 32) 864 ['input_1[0][0]']
bn_Conv1 (BatchNormalization) (None, 64, 64, 32) 128 ['Conv1[0][0]']
Conv1_relu (ReLU) (None, 64, 64, 32) 0 ['bn_Conv1[0][0]']
expanded_conv_depthwise (Depth (None, 64, 64, 32) 288 ['Conv1_relu[0][0]']
wiseConv2D)
expanded_conv_depthwise_BN (Ba (None, 64, 64, 32) 128 ['expanded_conv_depthwise[0][0]']
tchNormalization)
expanded_conv_depthwise_relu ( (None, 64, 64, 32) 0 ['expanded_conv_depthwise_BN[0][0
ReLU) ]']
expanded_conv_project (Conv2D) (None, 64, 64, 16) 512 ['expanded_conv_depthwise_relu[0]
[0]']
expanded_conv_project_BN (Batc (None, 64, 64, 16) 64 ['expanded_conv_project[0][0]']
hNormalization)
block_1_expand (Conv2D) (None, 64, 64, 96) 1536 ['expanded_conv_project_BN[0][0]'
]
block_1_expand_BN (BatchNormal (None, 64, 64, 96) 384 ['block_1_expand[0][0]']
ization)
block_1_expand_relu (ReLU) (None, 64, 64, 96) 0 ['block_1_expand_BN[0][0]']
block_1_pad (ZeroPadding2D) (None, 65, 65, 96) 0 ['block_1_expand_relu[0][0]']
block_1_depthwise (DepthwiseCo (None, 32, 32, 96) 864 ['block_1_pad[0][0]']
nv2D)
block_1_depthwise_BN (BatchNor (None, 32, 32, 96) 384 ['block_1_depthwise[0][0]']
malization)
block_1_depthwise_relu (ReLU) (None, 32, 32, 96) 0 ['block_1_depthwise_BN[0][0]']
block_1_project (Conv2D) (None, 32, 32, 24) 2304 ['block_1_depthwise_relu[0][0]']
block_1_project_BN (BatchNorma (None, 32, 32, 24) 96 ['block_1_project[0][0]']
lization)
block_2_expand (Conv2D) (None, 32, 32, 144) 3456 ['block_1_project_BN[0][0]']
block_2_expand_BN (BatchNormal (None, 32, 32, 144) 576 ['block_2_expand[0][0]']
ization)
block_2_expand_relu (ReLU) (None, 32, 32, 144) 0 ['block_2_expand_BN[0][0]']
block_2_depthwise (DepthwiseCo (None, 32, 32, 144) 1296 ['block_2_expand_relu[0][0]']
nv2D)
block_2_depthwise_BN (BatchNor (None, 32, 32, 144) 576 ['block_2_depthwise[0][0]']
malization)
block_2_depthwise_relu (ReLU) (None, 32, 32, 144) 0 ['block_2_depthwise_BN[0][0]']
block_2_project (Conv2D) (None, 32, 32, 24) 3456 ['block_2_depthwise_relu[0][0]']
block_2_project_BN (BatchNorma (None, 32, 32, 24) 96 ['block_2_project[0][0]']
lization)
block_2_add (Add) (None, 32, 32, 24) 0 ['block_1_project_BN[0][0]',
'block_2_project_BN[0][0]']
block_3_expand (Conv2D) (None, 32, 32, 144) 3456 ['block_2_add[0][0]']
block_3_expand_BN (BatchNormal (None, 32, 32, 144) 576 ['block_3_expand[0][0]']
ization)
block_3_expand_relu (ReLU) (None, 32, 32, 144) 0 ['block_3_expand_BN[0][0]']
block_3_pad (ZeroPadding2D) (None, 33, 33, 144) 0 ['block_3_expand_relu[0][0]']
block_3_depthwise (DepthwiseCo (None, 16, 16, 144) 1296 ['block_3_pad[0][0]']
nv2D)
block_3_depthwise_BN (BatchNor (None, 16, 16, 144) 576 ['block_3_depthwise[0][0]']
malization)
block_3_depthwise_relu (ReLU) (None, 16, 16, 144) 0 ['block_3_depthwise_BN[0][0]']
block_3_project (Conv2D) (None, 16, 16, 32) 4608 ['block_3_depthwise_relu[0][0]']
block_3_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_3_project[0][0]']
lization)
block_4_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_3_project_BN[0][0]']
block_4_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_4_expand[0][0]']
ization)
block_4_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_4_expand_BN[0][0]']
block_4_depthwise (DepthwiseCo (None, 16, 16, 192) 1728 ['block_4_expand_relu[0][0]']
nv2D)
block_4_depthwise_BN (BatchNor (None, 16, 16, 192) 768 ['block_4_depthwise[0][0]']
malization)
block_4_depthwise_relu (ReLU) (None, 16, 16, 192) 0 ['block_4_depthwise_BN[0][0]']
block_4_project (Conv2D) (None, 16, 16, 32) 6144 ['block_4_depthwise_relu[0][0]']
block_4_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_4_project[0][0]']
lization)
block_4_add (Add) (None, 16, 16, 32) 0 ['block_3_project_BN[0][0]',
'block_4_project_BN[0][0]']
block_5_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_4_add[0][0]']
block_5_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_5_expand[0][0]']
ization)
block_5_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_5_expand_BN[0][0]']
block_5_depthwise (DepthwiseCo (None, 16, 16, 192) 1728 ['block_5_expand_relu[0][0]']
nv2D)
block_5_depthwise_BN (BatchNor (None, 16, 16, 192) 768 ['block_5_depthwise[0][0]']
malization)
block_5_depthwise_relu (ReLU) (None, 16, 16, 192) 0 ['block_5_depthwise_BN[0][0]']
block_5_project (Conv2D) (None, 16, 16, 32) 6144 ['block_5_depthwise_relu[0][0]']
block_5_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_5_project[0][0]']
lization)
block_5_add (Add) (None, 16, 16, 32) 0 ['block_4_add[0][0]',
'block_5_project_BN[0][0]']
block_6_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_5_add[0][0]']
block_6_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_6_expand[0][0]']
ization)
block_6_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_6_expand_BN[0][0]']
block_6_pad (ZeroPadding2D) (None, 17, 17, 192) 0 ['block_6_expand_relu[0][0]']
block_6_depthwise (DepthwiseCo (None, 8, 8, 192) 1728 ['block_6_pad[0][0]']
nv2D)
block_6_depthwise_BN (BatchNor (None, 8, 8, 192) 768 ['block_6_depthwise[0][0]']
malization)
block_6_depthwise_relu (ReLU) (None, 8, 8, 192) 0 ['block_6_depthwise_BN[0][0]']
block_6_project (Conv2D) (None, 8, 8, 64) 12288 ['block_6_depthwise_relu[0][0]']
block_6_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_6_project[0][0]']
lization)
block_7_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_6_project_BN[0][0]']
block_7_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_7_expand[0][0]']
ization)
block_7_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_7_expand_BN[0][0]']
block_7_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_7_expand_relu[0][0]']
nv2D)
block_7_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_7_depthwise[0][0]']
malization)
block_7_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_7_depthwise_BN[0][0]']
block_7_project (Conv2D) (None, 8, 8, 64) 24576 ['block_7_depthwise_relu[0][0]']
block_7_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_7_project[0][0]']
lization)
block_7_add (Add) (None, 8, 8, 64) 0 ['block_6_project_BN[0][0]',
'block_7_project_BN[0][0]']
block_8_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_7_add[0][0]']
block_8_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_8_expand[0][0]']
ization)
block_8_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_8_expand_BN[0][0]']
block_8_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_8_expand_relu[0][0]']
nv2D)
block_8_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_8_depthwise[0][0]']
malization)
block_8_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_8_depthwise_BN[0][0]']
block_8_project (Conv2D) (None, 8, 8, 64) 24576 ['block_8_depthwise_relu[0][0]']
block_8_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_8_project[0][0]']
lization)
block_8_add (Add) (None, 8, 8, 64) 0 ['block_7_add[0][0]',
'block_8_project_BN[0][0]']
block_9_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_8_add[0][0]']
block_9_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_9_expand[0][0]']
ization)
block_9_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_9_expand_BN[0][0]']
block_9_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_9_expand_relu[0][0]']
nv2D)
block_9_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_9_depthwise[0][0]']
malization)
block_9_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_9_depthwise_BN[0][0]']
block_9_project (Conv2D) (None, 8, 8, 64) 24576 ['block_9_depthwise_relu[0][0]']
block_9_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_9_project[0][0]']
lization)
block_9_add (Add) (None, 8, 8, 64) 0 ['block_8_add[0][0]',
'block_9_project_BN[0][0]']
block_10_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_9_add[0][0]']
block_10_expand_BN (BatchNorma (None, 8, 8, 384) 1536 ['block_10_expand[0][0]']
lization)
block_10_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_10_expand_BN[0][0]']
block_10_depthwise (DepthwiseC (None, 8, 8, 384) 3456 ['block_10_expand_relu[0][0]']
onv2D)
block_10_depthwise_BN (BatchNo (None, 8, 8, 384) 1536 ['block_10_depthwise[0][0]']
rmalization)
block_10_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_10_depthwise_BN[0][0]']
block_10_project (Conv2D) (None, 8, 8, 96) 36864 ['block_10_depthwise_relu[0][0]']
block_10_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_10_project[0][0]']
alization)
block_11_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_10_project_BN[0][0]']
block_11_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_11_expand[0][0]']
lization)
block_11_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_11_expand_BN[0][0]']
block_11_depthwise (DepthwiseC (None, 8, 8, 576) 5184 ['block_11_expand_relu[0][0]']
onv2D)
block_11_depthwise_BN (BatchNo (None, 8, 8, 576) 2304 ['block_11_depthwise[0][0]']
rmalization)
block_11_depthwise_relu (ReLU) (None, 8, 8, 576) 0 ['block_11_depthwise_BN[0][0]']
block_11_project (Conv2D) (None, 8, 8, 96) 55296 ['block_11_depthwise_relu[0][0]']
block_11_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_11_project[0][0]']
alization)
block_11_add (Add) (None, 8, 8, 96) 0 ['block_10_project_BN[0][0]',
'block_11_project_BN[0][0]']
block_12_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_11_add[0][0]']
block_12_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_12_expand[0][0]']
lization)
block_12_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_12_expand_BN[0][0]']
block_12_depthwise (DepthwiseC (None, 8, 8, 576) 5184 ['block_12_expand_relu[0][0]']
onv2D)
block_12_depthwise_BN (BatchNo (None, 8, 8, 576) 2304 ['block_12_depthwise[0][0]']
rmalization)
block_12_depthwise_relu (ReLU) (None, 8, 8, 576) 0 ['block_12_depthwise_BN[0][0]']
block_12_project (Conv2D) (None, 8, 8, 96) 55296 ['block_12_depthwise_relu[0][0]']
block_12_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_12_project[0][0]']
alization)
block_12_add (Add) (None, 8, 8, 96) 0 ['block_11_add[0][0]',
'block_12_project_BN[0][0]']
block_13_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_12_add[0][0]']
block_13_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_13_expand[0][0]']
lization)
block_13_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_13_expand_BN[0][0]']
block_13_pad (ZeroPadding2D) (None, 9, 9, 576) 0 ['block_13_expand_relu[0][0]']
block_13_depthwise (DepthwiseC (None, 4, 4, 576) 5184 ['block_13_pad[0][0]']
onv2D)
block_13_depthwise_BN (BatchNo (None, 4, 4, 576) 2304 ['block_13_depthwise[0][0]']
rmalization)
block_13_depthwise_relu (ReLU) (None, 4, 4, 576) 0 ['block_13_depthwise_BN[0][0]']
block_13_project (Conv2D) (None, 4, 4, 160) 92160 ['block_13_depthwise_relu[0][0]']
block_13_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_13_project[0][0]']
alization)
block_14_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_13_project_BN[0][0]']
block_14_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_14_expand[0][0]']
lization)
block_14_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_14_expand_BN[0][0]']
block_14_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_14_expand_relu[0][0]']
onv2D)
block_14_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_14_depthwise[0][0]']
rmalization)
block_14_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_14_depthwise_BN[0][0]']
block_14_project (Conv2D) (None, 4, 4, 160) 153600 ['block_14_depthwise_relu[0][0]']
block_14_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_14_project[0][0]']
alization)
block_14_add (Add) (None, 4, 4, 160) 0 ['block_13_project_BN[0][0]',
'block_14_project_BN[0][0]']
block_15_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_14_add[0][0]']
block_15_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_15_expand[0][0]']
lization)
block_15_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_15_expand_BN[0][0]']
block_15_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_15_expand_relu[0][0]']
onv2D)
block_15_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_15_depthwise[0][0]']
rmalization)
block_15_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_15_depthwise_BN[0][0]']
block_15_project (Conv2D) (None, 4, 4, 160) 153600 ['block_15_depthwise_relu[0][0]']
block_15_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_15_project[0][0]']
alization)
block_15_add (Add) (None, 4, 4, 160) 0 ['block_14_add[0][0]',
'block_15_project_BN[0][0]']
block_16_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_15_add[0][0]']
block_16_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_16_expand[0][0]']
lization)
block_16_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_16_expand_BN[0][0]']
block_16_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_16_expand_relu[0][0]']
onv2D)
block_16_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_16_depthwise[0][0]']
rmalization)
block_16_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_16_depthwise_BN[0][0]']
block_16_project (Conv2D) (None, 4, 4, 320) 307200 ['block_16_depthwise_relu[0][0]']
block_16_project_BN (BatchNorm (None, 4, 4, 320) 1280 ['block_16_project[0][0]']
alization)
Conv_1 (Conv2D) (None, 4, 4, 1280) 409600 ['block_16_project_BN[0][0]']
Conv_1_bn (BatchNormalization) (None, 4, 4, 1280) 5120 ['Conv_1[0][0]']
out_relu (ReLU) (None, 4, 4, 1280) 0 ['Conv_1_bn[0][0]']
flatten (Flatten) (None, 20480) 0 ['out_relu[0][0]']
dense (Dense) (None, 128) 2621568 ['flatten[0][0]']
dense_1 (Dense) (None, 1) 129 ['dense[0][0]']
==================================================================================================
Total params: 4,879,681
Trainable params: 2,621,697
Non-trainable params: 2,257,984
__________________________________________________________________________________________________
Compiling the model¶
In [30]:
from keras.optimizers import RMSprop
In [31]:
model.compile(RMSprop(0.0001), 'binary_crossentropy', ['accuracy'] )
In [ ]:
Creating Data Generators¶
Resizing images
Big pre-trained architecture support only certain input sizes.
For example: MobileNet (architecture that we use) supports: (96, 96), (128, 128), (160, 160), (192, 192), (224, 224).
In [32]:
# 실무에서는, 데이터 증강까지 하도록 한다.
train_datagen = ImageDataGenerator(rescale = 1/255.0, width_shift_range=0.2 ) # 넘파이로 만들기 // 시간문제로 증강은 하나만하였음
In [33]:
test_datagen = ImageDataGenerator(rescale = 1/255.0 )
In [34]:
train_generator = train_datagen.flow_from_directory( train_dir, target_size = (128,128),
class_mode = 'binary',
batch_size=128) # 디렉토리에서 읽어와라 // batch_size=128 == 128개씩 학습 // class_mode = 'binary' == 디렉토리를 자동으로 2개로 레이블링함.
Found 2000 images belonging to 2 classes.
In [35]:
test_generator = test_datagen.flow_from_directory( test_dir, target_size = (128,128),
class_mode = 'binary',
batch_size=128)
Found 1000 images belonging to 2 classes.
Training the model¶
In [36]:
epoch_history = model.fit(train_generator, epochs = 5 , validation_data = (test_generator) )
Epoch 1/5 16/16 [==============================] - 20s 638ms/step - loss: 0.2892 - accuracy: 0.8795 - val_loss: 0.1138 - val_accuracy: 0.9520 Epoch 2/5 16/16 [==============================] - 8s 522ms/step - loss: 0.0504 - accuracy: 0.9825 - val_loss: 0.1253 - val_accuracy: 0.9480 Epoch 3/5 16/16 [==============================] - 8s 525ms/step - loss: 0.0244 - accuracy: 0.9950 - val_loss: 0.0867 - val_accuracy: 0.9660 Epoch 4/5 16/16 [==============================] - 10s 626ms/step - loss: 0.0594 - accuracy: 0.9835 - val_loss: 0.0813 - val_accuracy: 0.9750 Epoch 5/5 16/16 [==============================] - 8s 526ms/step - loss: 0.0039 - accuracy: 1.0000 - val_loss: 0.0814 - val_accuracy: 0.9750
In [ ]:
Transfer learning model evaluation¶
In [37]:
model.evaluate(test_generator)
8/8 [==============================] - 3s 344ms/step - loss: 0.0814 - accuracy: 0.9750
Out[37]:
[0.08142396807670593, 0.9750000238418579]
In [ ]:
In [ ]:
Fine tuning¶
There are a few pointers:
- DO NOT use Fine tuning on the whole network; only a few top layers are enough. In most cases, they are more specialized. The goal of the Fine-tuning is to adopt that specific part of the network for our custom (new) dataset.
- Start with the fine tunning AFTER you have finished with transfer learning step. If we try to perform Fine tuning immediately, gradients will be much different between our custom head layer and a few unfrozen layers from the base model.
In [ ]:
# 파인 튜닝 : 섬세한 튜닝
# 이 방법은, 꼭 !!! 트랜스퍼 러닝을 한 다음에, 수행하는 방법이다.
# 위의 트랜스퍼 러닝을 한 후에, 조금 더 개선이 가능한지 해보는 방법으로서
# 위에서 학습한 모델을 가지고!!! 그 상태에서, 추가로 학습을 시키는것!
# 단, 좋은 모델의 일부분을, 우리 데이터로 학습가능하도록 변경한 후에 학습시킨다.
Un-freeze a few top layers from the model¶
In [ ]:
# 1. 일단은, 베이스모델의 전체 레이어를 다시, 학습 가능토록 먼저 바꿔준다.
In [38]:
base_model.trainable = True
In [40]:
# 2. 베이스 모델의 전체 레이어 수를 확인한다.
len( base_model.layers )
Out[40]:
154
In [42]:
# 3. 위에서, 레이어 갯수를 확인했으니,
# 몇번째 레이어까지 학습이 안되도록 할것인지, 결정해 준다.
end_layer = 100 # 정답이없다, 실험해야함
In [44]:
for layer in base_model.layers[ 0 : end_layer+1 ] :
layer.trainable = False
In [45]:
base_model.summary()
Model: "mobilenetv2_1.00_128"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 128, 128, 3 0 []
)]
Conv1 (Conv2D) (None, 64, 64, 32) 864 ['input_1[0][0]']
bn_Conv1 (BatchNormalization) (None, 64, 64, 32) 128 ['Conv1[0][0]']
Conv1_relu (ReLU) (None, 64, 64, 32) 0 ['bn_Conv1[0][0]']
expanded_conv_depthwise (Depth (None, 64, 64, 32) 288 ['Conv1_relu[0][0]']
wiseConv2D)
expanded_conv_depthwise_BN (Ba (None, 64, 64, 32) 128 ['expanded_conv_depthwise[0][0]']
tchNormalization)
expanded_conv_depthwise_relu ( (None, 64, 64, 32) 0 ['expanded_conv_depthwise_BN[0][0
ReLU) ]']
expanded_conv_project (Conv2D) (None, 64, 64, 16) 512 ['expanded_conv_depthwise_relu[0]
[0]']
expanded_conv_project_BN (Batc (None, 64, 64, 16) 64 ['expanded_conv_project[0][0]']
hNormalization)
block_1_expand (Conv2D) (None, 64, 64, 96) 1536 ['expanded_conv_project_BN[0][0]'
]
block_1_expand_BN (BatchNormal (None, 64, 64, 96) 384 ['block_1_expand[0][0]']
ization)
block_1_expand_relu (ReLU) (None, 64, 64, 96) 0 ['block_1_expand_BN[0][0]']
block_1_pad (ZeroPadding2D) (None, 65, 65, 96) 0 ['block_1_expand_relu[0][0]']
block_1_depthwise (DepthwiseCo (None, 32, 32, 96) 864 ['block_1_pad[0][0]']
nv2D)
block_1_depthwise_BN (BatchNor (None, 32, 32, 96) 384 ['block_1_depthwise[0][0]']
malization)
block_1_depthwise_relu (ReLU) (None, 32, 32, 96) 0 ['block_1_depthwise_BN[0][0]']
block_1_project (Conv2D) (None, 32, 32, 24) 2304 ['block_1_depthwise_relu[0][0]']
block_1_project_BN (BatchNorma (None, 32, 32, 24) 96 ['block_1_project[0][0]']
lization)
block_2_expand (Conv2D) (None, 32, 32, 144) 3456 ['block_1_project_BN[0][0]']
block_2_expand_BN (BatchNormal (None, 32, 32, 144) 576 ['block_2_expand[0][0]']
ization)
block_2_expand_relu (ReLU) (None, 32, 32, 144) 0 ['block_2_expand_BN[0][0]']
block_2_depthwise (DepthwiseCo (None, 32, 32, 144) 1296 ['block_2_expand_relu[0][0]']
nv2D)
block_2_depthwise_BN (BatchNor (None, 32, 32, 144) 576 ['block_2_depthwise[0][0]']
malization)
block_2_depthwise_relu (ReLU) (None, 32, 32, 144) 0 ['block_2_depthwise_BN[0][0]']
block_2_project (Conv2D) (None, 32, 32, 24) 3456 ['block_2_depthwise_relu[0][0]']
block_2_project_BN (BatchNorma (None, 32, 32, 24) 96 ['block_2_project[0][0]']
lization)
block_2_add (Add) (None, 32, 32, 24) 0 ['block_1_project_BN[0][0]',
'block_2_project_BN[0][0]']
block_3_expand (Conv2D) (None, 32, 32, 144) 3456 ['block_2_add[0][0]']
block_3_expand_BN (BatchNormal (None, 32, 32, 144) 576 ['block_3_expand[0][0]']
ization)
block_3_expand_relu (ReLU) (None, 32, 32, 144) 0 ['block_3_expand_BN[0][0]']
block_3_pad (ZeroPadding2D) (None, 33, 33, 144) 0 ['block_3_expand_relu[0][0]']
block_3_depthwise (DepthwiseCo (None, 16, 16, 144) 1296 ['block_3_pad[0][0]']
nv2D)
block_3_depthwise_BN (BatchNor (None, 16, 16, 144) 576 ['block_3_depthwise[0][0]']
malization)
block_3_depthwise_relu (ReLU) (None, 16, 16, 144) 0 ['block_3_depthwise_BN[0][0]']
block_3_project (Conv2D) (None, 16, 16, 32) 4608 ['block_3_depthwise_relu[0][0]']
block_3_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_3_project[0][0]']
lization)
block_4_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_3_project_BN[0][0]']
block_4_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_4_expand[0][0]']
ization)
block_4_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_4_expand_BN[0][0]']
block_4_depthwise (DepthwiseCo (None, 16, 16, 192) 1728 ['block_4_expand_relu[0][0]']
nv2D)
block_4_depthwise_BN (BatchNor (None, 16, 16, 192) 768 ['block_4_depthwise[0][0]']
malization)
block_4_depthwise_relu (ReLU) (None, 16, 16, 192) 0 ['block_4_depthwise_BN[0][0]']
block_4_project (Conv2D) (None, 16, 16, 32) 6144 ['block_4_depthwise_relu[0][0]']
block_4_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_4_project[0][0]']
lization)
block_4_add (Add) (None, 16, 16, 32) 0 ['block_3_project_BN[0][0]',
'block_4_project_BN[0][0]']
block_5_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_4_add[0][0]']
block_5_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_5_expand[0][0]']
ization)
block_5_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_5_expand_BN[0][0]']
block_5_depthwise (DepthwiseCo (None, 16, 16, 192) 1728 ['block_5_expand_relu[0][0]']
nv2D)
block_5_depthwise_BN (BatchNor (None, 16, 16, 192) 768 ['block_5_depthwise[0][0]']
malization)
block_5_depthwise_relu (ReLU) (None, 16, 16, 192) 0 ['block_5_depthwise_BN[0][0]']
block_5_project (Conv2D) (None, 16, 16, 32) 6144 ['block_5_depthwise_relu[0][0]']
block_5_project_BN (BatchNorma (None, 16, 16, 32) 128 ['block_5_project[0][0]']
lization)
block_5_add (Add) (None, 16, 16, 32) 0 ['block_4_add[0][0]',
'block_5_project_BN[0][0]']
block_6_expand (Conv2D) (None, 16, 16, 192) 6144 ['block_5_add[0][0]']
block_6_expand_BN (BatchNormal (None, 16, 16, 192) 768 ['block_6_expand[0][0]']
ization)
block_6_expand_relu (ReLU) (None, 16, 16, 192) 0 ['block_6_expand_BN[0][0]']
block_6_pad (ZeroPadding2D) (None, 17, 17, 192) 0 ['block_6_expand_relu[0][0]']
block_6_depthwise (DepthwiseCo (None, 8, 8, 192) 1728 ['block_6_pad[0][0]']
nv2D)
block_6_depthwise_BN (BatchNor (None, 8, 8, 192) 768 ['block_6_depthwise[0][0]']
malization)
block_6_depthwise_relu (ReLU) (None, 8, 8, 192) 0 ['block_6_depthwise_BN[0][0]']
block_6_project (Conv2D) (None, 8, 8, 64) 12288 ['block_6_depthwise_relu[0][0]']
block_6_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_6_project[0][0]']
lization)
block_7_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_6_project_BN[0][0]']
block_7_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_7_expand[0][0]']
ization)
block_7_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_7_expand_BN[0][0]']
block_7_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_7_expand_relu[0][0]']
nv2D)
block_7_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_7_depthwise[0][0]']
malization)
block_7_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_7_depthwise_BN[0][0]']
block_7_project (Conv2D) (None, 8, 8, 64) 24576 ['block_7_depthwise_relu[0][0]']
block_7_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_7_project[0][0]']
lization)
block_7_add (Add) (None, 8, 8, 64) 0 ['block_6_project_BN[0][0]',
'block_7_project_BN[0][0]']
block_8_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_7_add[0][0]']
block_8_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_8_expand[0][0]']
ization)
block_8_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_8_expand_BN[0][0]']
block_8_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_8_expand_relu[0][0]']
nv2D)
block_8_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_8_depthwise[0][0]']
malization)
block_8_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_8_depthwise_BN[0][0]']
block_8_project (Conv2D) (None, 8, 8, 64) 24576 ['block_8_depthwise_relu[0][0]']
block_8_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_8_project[0][0]']
lization)
block_8_add (Add) (None, 8, 8, 64) 0 ['block_7_add[0][0]',
'block_8_project_BN[0][0]']
block_9_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_8_add[0][0]']
block_9_expand_BN (BatchNormal (None, 8, 8, 384) 1536 ['block_9_expand[0][0]']
ization)
block_9_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_9_expand_BN[0][0]']
block_9_depthwise (DepthwiseCo (None, 8, 8, 384) 3456 ['block_9_expand_relu[0][0]']
nv2D)
block_9_depthwise_BN (BatchNor (None, 8, 8, 384) 1536 ['block_9_depthwise[0][0]']
malization)
block_9_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_9_depthwise_BN[0][0]']
block_9_project (Conv2D) (None, 8, 8, 64) 24576 ['block_9_depthwise_relu[0][0]']
block_9_project_BN (BatchNorma (None, 8, 8, 64) 256 ['block_9_project[0][0]']
lization)
block_9_add (Add) (None, 8, 8, 64) 0 ['block_8_add[0][0]',
'block_9_project_BN[0][0]']
block_10_expand (Conv2D) (None, 8, 8, 384) 24576 ['block_9_add[0][0]']
block_10_expand_BN (BatchNorma (None, 8, 8, 384) 1536 ['block_10_expand[0][0]']
lization)
block_10_expand_relu (ReLU) (None, 8, 8, 384) 0 ['block_10_expand_BN[0][0]']
block_10_depthwise (DepthwiseC (None, 8, 8, 384) 3456 ['block_10_expand_relu[0][0]']
onv2D)
block_10_depthwise_BN (BatchNo (None, 8, 8, 384) 1536 ['block_10_depthwise[0][0]']
rmalization)
block_10_depthwise_relu (ReLU) (None, 8, 8, 384) 0 ['block_10_depthwise_BN[0][0]']
block_10_project (Conv2D) (None, 8, 8, 96) 36864 ['block_10_depthwise_relu[0][0]']
block_10_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_10_project[0][0]']
alization)
block_11_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_10_project_BN[0][0]']
block_11_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_11_expand[0][0]']
lization)
block_11_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_11_expand_BN[0][0]']
block_11_depthwise (DepthwiseC (None, 8, 8, 576) 5184 ['block_11_expand_relu[0][0]']
onv2D)
block_11_depthwise_BN (BatchNo (None, 8, 8, 576) 2304 ['block_11_depthwise[0][0]']
rmalization)
block_11_depthwise_relu (ReLU) (None, 8, 8, 576) 0 ['block_11_depthwise_BN[0][0]']
block_11_project (Conv2D) (None, 8, 8, 96) 55296 ['block_11_depthwise_relu[0][0]']
block_11_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_11_project[0][0]']
alization)
block_11_add (Add) (None, 8, 8, 96) 0 ['block_10_project_BN[0][0]',
'block_11_project_BN[0][0]']
block_12_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_11_add[0][0]']
block_12_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_12_expand[0][0]']
lization)
block_12_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_12_expand_BN[0][0]']
block_12_depthwise (DepthwiseC (None, 8, 8, 576) 5184 ['block_12_expand_relu[0][0]']
onv2D)
block_12_depthwise_BN (BatchNo (None, 8, 8, 576) 2304 ['block_12_depthwise[0][0]']
rmalization)
block_12_depthwise_relu (ReLU) (None, 8, 8, 576) 0 ['block_12_depthwise_BN[0][0]']
block_12_project (Conv2D) (None, 8, 8, 96) 55296 ['block_12_depthwise_relu[0][0]']
block_12_project_BN (BatchNorm (None, 8, 8, 96) 384 ['block_12_project[0][0]']
alization)
block_12_add (Add) (None, 8, 8, 96) 0 ['block_11_add[0][0]',
'block_12_project_BN[0][0]']
block_13_expand (Conv2D) (None, 8, 8, 576) 55296 ['block_12_add[0][0]']
block_13_expand_BN (BatchNorma (None, 8, 8, 576) 2304 ['block_13_expand[0][0]']
lization)
block_13_expand_relu (ReLU) (None, 8, 8, 576) 0 ['block_13_expand_BN[0][0]']
block_13_pad (ZeroPadding2D) (None, 9, 9, 576) 0 ['block_13_expand_relu[0][0]']
block_13_depthwise (DepthwiseC (None, 4, 4, 576) 5184 ['block_13_pad[0][0]']
onv2D)
block_13_depthwise_BN (BatchNo (None, 4, 4, 576) 2304 ['block_13_depthwise[0][0]']
rmalization)
block_13_depthwise_relu (ReLU) (None, 4, 4, 576) 0 ['block_13_depthwise_BN[0][0]']
block_13_project (Conv2D) (None, 4, 4, 160) 92160 ['block_13_depthwise_relu[0][0]']
block_13_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_13_project[0][0]']
alization)
block_14_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_13_project_BN[0][0]']
block_14_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_14_expand[0][0]']
lization)
block_14_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_14_expand_BN[0][0]']
block_14_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_14_expand_relu[0][0]']
onv2D)
block_14_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_14_depthwise[0][0]']
rmalization)
block_14_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_14_depthwise_BN[0][0]']
block_14_project (Conv2D) (None, 4, 4, 160) 153600 ['block_14_depthwise_relu[0][0]']
block_14_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_14_project[0][0]']
alization)
block_14_add (Add) (None, 4, 4, 160) 0 ['block_13_project_BN[0][0]',
'block_14_project_BN[0][0]']
block_15_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_14_add[0][0]']
block_15_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_15_expand[0][0]']
lization)
block_15_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_15_expand_BN[0][0]']
block_15_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_15_expand_relu[0][0]']
onv2D)
block_15_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_15_depthwise[0][0]']
rmalization)
block_15_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_15_depthwise_BN[0][0]']
block_15_project (Conv2D) (None, 4, 4, 160) 153600 ['block_15_depthwise_relu[0][0]']
block_15_project_BN (BatchNorm (None, 4, 4, 160) 640 ['block_15_project[0][0]']
alization)
block_15_add (Add) (None, 4, 4, 160) 0 ['block_14_add[0][0]',
'block_15_project_BN[0][0]']
block_16_expand (Conv2D) (None, 4, 4, 960) 153600 ['block_15_add[0][0]']
block_16_expand_BN (BatchNorma (None, 4, 4, 960) 3840 ['block_16_expand[0][0]']
lization)
block_16_expand_relu (ReLU) (None, 4, 4, 960) 0 ['block_16_expand_BN[0][0]']
block_16_depthwise (DepthwiseC (None, 4, 4, 960) 8640 ['block_16_expand_relu[0][0]']
onv2D)
block_16_depthwise_BN (BatchNo (None, 4, 4, 960) 3840 ['block_16_depthwise[0][0]']
rmalization)
block_16_depthwise_relu (ReLU) (None, 4, 4, 960) 0 ['block_16_depthwise_BN[0][0]']
block_16_project (Conv2D) (None, 4, 4, 320) 307200 ['block_16_depthwise_relu[0][0]']
block_16_project_BN (BatchNorm (None, 4, 4, 320) 1280 ['block_16_project[0][0]']
alization)
Conv_1 (Conv2D) (None, 4, 4, 1280) 409600 ['block_16_project_BN[0][0]']
Conv_1_bn (BatchNormalization) (None, 4, 4, 1280) 5120 ['Conv_1[0][0]']
out_relu (ReLU) (None, 4, 4, 1280) 0 ['Conv_1_bn[0][0]']
==================================================================================================
Total params: 2,257,984
Trainable params: 1,861,440
Non-trainable params: 396,544
__________________________________________________________________________________________________
In [ ]:
In [ ]:
Compiling the model for fine-tuning¶
In [46]:
from keras.optimizers import Adam
In [47]:
model.compile(Adam(0.0001) , 'binary_crossentropy', ['accuracy']) # 러닝레잇도 찾아내야한다(실험으로) , 옵티마이저도 변경이 가능.
In [ ]:
Fine tuning¶
In [48]:
epoch_history2 = model.fit(train_generator, epochs=5, validation_data = (test_generator))
Epoch 1/5 16/16 [==============================] - 14s 589ms/step - loss: 0.1623 - accuracy: 0.9325 - val_loss: 0.2928 - val_accuracy: 0.9230 Epoch 2/5 16/16 [==============================] - 9s 545ms/step - loss: 0.0159 - accuracy: 0.9950 - val_loss: 0.1181 - val_accuracy: 0.9670 Epoch 3/5 16/16 [==============================] - 9s 539ms/step - loss: 0.0046 - accuracy: 0.9995 - val_loss: 0.1553 - val_accuracy: 0.9630 Epoch 4/5 16/16 [==============================] - 9s 543ms/step - loss: 0.0023 - accuracy: 1.0000 - val_loss: 0.1728 - val_accuracy: 0.9610 Epoch 5/5 16/16 [==============================] - 8s 536ms/step - loss: 8.8268e-04 - accuracy: 1.0000 - val_loss: 0.1562 - val_accuracy: 0.9660
Evaluating the fine tuned model¶
In [49]:
model.evaluate(test_generator)
8/8 [==============================] - 3s 345ms/step - loss: 0.1562 - accuracy: 0.9660
Out[49]:
[0.1562417596578598, 0.9660000205039978]
In [ ]:
In [ ]:
In [ ]:
In [ ]:
이미지파일 테스트 해본다.¶
'DataScience > TensorFlow[CNN]' 카테고리의 다른 글
| 딥러닝 텐서플로우 에포크 시마다, 가장 좋은 모델을 저장하는, ModelCheckpoint 사용방법, 에포크 시마다, 기록을 남길 수 있는, CSVLogger 사용 방법 (0) | 2023.01.02 |
|---|---|
| 딥러니 텐서플로우 CNN을 이용하여, CIFAR-10 이미지 분류하기, 데이트셋이 이미 넘파이라면? (0) | 2023.01.02 |
| 딥러닝 텐서플로우 이미지 증강 예시 (0) | 2022.12.30 |
| 딥러닝 텐서플로우 개와 고양이 분류하는 Neural Network 만들기 (0) | 2022.12.30 |
| 딥러닝 텐서플로우 CNN 인간,말 분류, 파이썬으로 압축풀기, 이미지파일을 넘파이 어레이로가져오기 (0) | 2022.12.30 |