问题描述
我知道ImageDataGenerator
为每个输入图像生成一个随机扩充的图像。现在,我想为每个输入图像生成两个增强图像:
datagen = tf.keras.preprocessing.image.ImageDataGenerator(
rotation_range=40,width_shift_range=0.2,height_shift_range=0.2,shear_range=0.2,zoom_range=0.2,horizontal_flip=True,fill_mode='nearest')
train_ds = datagen.flow_from_directory('/home/train/')
为进一步说明,我想在同一张图像上应用2个不同的增强函数,即,如果我们对5张图像进行采样,最终将在批处理中得到2×5 = 10个增强观察
那么我该如何进行?
解决方法
我建议创建一个从tf.keras.utils.Sequence继承的自定义数据生成器。有很多方法可以解决此问题,但这应该符合您的寻找思路:
class double_aug_generator(tf.keras.utils.Sequence):
def __init__(self,x,y,batch_size,aug_params1,aug_params2):
self.x,self.y = x,y
self.batch_size = batch_size
self.datagen = tf.keras.preprocessing.image.ImageDataGenerator(**aug_params1)
// dictionary of parameters for the second augmentation
self.aug_params2 = aug_params2
def __len__(self):
return math.ceil(len(self.x) / self.batch_size)
def load(self,file_names):
// load and return raw images however you like
def __getitem__(self,idx):
batch_x = self.x[idx * self.batch_size:(idx + 1) *
self.batch_size]
batch_y = self.y[idx * self.batch_size:(idx + 1) *
self.batch_size]
// load images
batch_x = self.load(batch_x)
// apply first augmentation
batch_x = self.datagen.flow(batch_x)
// apply second
batch_x = self.datagen.apply_transform(batch_x,self.aug_params2)
return batch_x,np.array(batch_y)