dogbreed_keras_mobilenet.py

import os
import numpy as np

from keras.applications.mobilenet import MobileNet
from keras.models import Sequential, Model
from keras.layers import Input, Dense, Activation, GlobalAveragePooling2D, Reshape, Conv2D, Dropout
from keras.optimizers import adam
from keras.preprocessing.image import ImageDataGenerator
from keras.callbacks import ModelCheckpoint, TensorBoard, ReduceLROnPlateau, EarlyStopping
from keras.backend import backend
from keras.utils import plot_model

import tensorflow as tf

from numpy.random import seed
from tensorflow import set_random_seed
seed(1)
set_random_seed(2)


config = tf.ConfigProto()
config.gpu_options.allow_growth=True
sess = tf.Session(config=config)

img_size = 224
batch_size = 32
num_epochs = 20
num_classes = 120


# def get_class_weights(dir_path):
#     class_names = os.listdir(dir_path)
#     class_names.sort()
#     class_el_nums = []
#     for class_name in range(num_classes):
#         filenames = os.listdir(os.path.join(dir_path, str(class_name)))
#         class_el_nums.append(len(filenames))
#     class_weights = []
#     for class_idx in range(num_classes):
#         if class_el_nums[class_idx] == 0:
#             class_weights.append(1.0)
#         else:
#             class_weights.append(1. / class_el_nums[class_idx])

#     return class_weights

base_model = MobileNet(input_shape=(img_size, img_size, 3), include_top=False)
# base_model = MobileNet(input_shape=(img_size, img_size, 3))
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(1024, activation='relu')(x)
predictions = Dense(num_classes, activation='softmax')(x)

model = Model(input=base_model.input, output=predictions)
# model.summary()
# plot_model(model, 'model.png', show_shapes = True)

for layer in base_model.layers:
    layer.trainable = False

model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['acc'])

train_datagen = ImageDataGenerator(
    rescale=1./255,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True
)

test_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory(
    './data/train/',
    target_size=(img_size, img_size),
    batch_size=batch_size
)

val_generator = test_datagen.flow_from_directory(
    './data/val/',
    target_size=(img_size, img_size),
    batch_size=batch_size
)

prefix = "1_w"
filepath = "./models/" + prefix + "_{epoch:02d}_{val_acc:.2f}.hdf5"
os.makedirs("./models", exist_ok=True)
checkpoint = ModelCheckpoint(filepath, monitor='val_acc', verbose=1, save_best_only=True, mode='max')
tensorboard = TensorBoard(log_dir='./logs/' + prefix)
reduce_lr = ReduceLROnPlateau(monitor='val_acc', mode = 'max',factor=0.2, patience=5, min_lr=0.00001, verbose=1)
early_stopping = EarlyStopping(monitor='val_acc', mode = 'max',patience=15, verbose=1)
callbacks_list = [checkpoint, tensorboard, reduce_lr, early_stopping]

model.fit_generator(
    train_generator,
    steps_per_epoch=40000 // batch_size,
    epochs=num_epochs,
    callbacks=callbacks_list,
    validation_data=val_generator,
    validation_steps=8000 // batch_size
)