【发布时间】:2020-06-15 19:20:10
【问题描述】:
以下是我正在尝试使用的示例代码。我想知道是否有任何方法可以在应用 imagedatagenerator() 之后和执行训练(即 .fit_generator)之前立即获取图像数量。原因是,我想稍后使用这些图像而不是我的原始数据集进行训练。
train_datagen=ImageDataGenerator(rescale=1./255,
#featurewise_center=True,
samplewise_center=True,
zca_epsilon=1e-06,
#channel_shift_range=100.0,
#samplewise_std_normalization=True,
#featurewise_std_normalization=True,
rotation_range=15,
#width_shift_range=0,
#height_shift_range=0,
shear_range=0.2,
fill_mode='nearest',
zoom_range=0.1,
horizontal_flip= True,
)
val_datagen= ImageDataGenerator(rescale=1./255,
samplewise_center=True,
)
train_generator= train_datagen.flow(X_train, Y_train, batch_size=batch_size,shuffle=True)
val_generator= val_datagen.flow(X_val, Y_val,batch_size=batch_size,shuffle=True)
history= model.fit_generator(train_generator,
batch_size= batch_size,
steps_per_epoch=trainSize,
epochs=10,
validation_data=val_generator,
validation_steps=valSize,
callbacks=[LearningRateScheduler(lr_schedule)]
#callbacks=[es_callback]
)
【问题讨论】:
标签: tensorflow image-processing keras conv-neural-network