Question

我正在尝试使用VGG16预训练模型创建一个非常简单的CNN程序进行分类。我的数据集是在Kaggle上发现的第一代神奇宝贝，它包含149个不同类别的10.000张图像。问题是我没有得到足够的准确性，我可以得到的最大值几乎是40％。

代码如下：

import tensorflow as tf
import numpy as np

vgg_model = tf.keras.applications.VGG16(weights='imagenet', include_top=False, input_shape = (224,224,3))
vgg_model.trainable = False

model = tf.keras.models.Sequential()
model.add(vgg_model)
model.add(tf.keras.layers.Flatten(input_shape=vgg_model.output_shape[1:]))
model.add(tf.keras.layers.Dense(256, activation='relu'))
model.add(tf.keras.layers.Dense(149, activation='softmax'))

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

train= tf.keras.preprocessing.image.ImageDataGenerator(rescale=1./255, shear_range=0.2, zoom_range=0.2, horizontal_flip=True)
test= tf.keras.preprocessing.image.ImageDataGenerator(rescale=1./255)
training_set = train.flow_from_directory('datasets/generation/train', target_size=(224,224), class_mode = 'categorical')
val_set = train.flow_from_directory('datasets/generation/test', target_size=(224,224), class_mode = 'categorical')

history = model.fit_generator(training_set, steps_per_epoch = 64, epochs = 30, validation_data = val_set, validation_steps = 64)

以下是输出：

Epoch 1/30
64/64 [=================] - 57s 885ms/step - loss: 5.0538 - acc: 0.0410 - val_loss: 4.7750 - val_acc: 0.0659
Epoch 2/30
64/64 [=================] - 50s 775ms/step - loss: 4.7432 - acc: 0.0747 - val_loss: 4.5880 - val_acc: 0.1037
...
Epoch 10/30
64/64 [=================] - 50s 788ms/step - loss: 3.0594 - acc: 0.3077 - val_loss: 3,3569 - val_acc: 0.2425
...
Epoch 20/30
64/64 [=================] - 54s 843ms/step - loss: 2.2030 - acc: 0.4628 - val_loss: 2.8968 - val_acc: 0.3565
...
Epoch 25/30
64/64 [=================] - 49s 773ms/step - loss: 1.9324 - acc: 0.5293 - val_loss: 2.6801 - val_acc: 0.3823
...
Epoch 30/30
64/64 [=================] - 52s 814ms/step - loss: 1.6427 - acc: 0.5801 - val_loss: 2.6852 - val_acc: 0.3936

有人可以帮助我理解吗？

Answer 1

这可以提高您的验证准确性：

# Add a fully connected layer with 512 hidden units and ReLU activation
x = keras.layers.Dense(512, activation='relu')(x)

# Add a dropout rate of 0.5
x = keras.layers.Dropout(0.5)(x)

Keras预训练的VGG16低精度

1 个答案: