Tensorflow数据增强

Question

我想转换这个keras数据增强工作流：

datagen = ImageDataGenerator( 
    rescale=1./255,
    rotation_range = 10,
    horizontal_flip = True,
    width_shift_range=0.1,
    height_shift_range=0.1,
    fill_mode = 'nearest')

这是一个代码片段，但这两个函数都不起作用，因为它不支持批处理维度！

import numpy as np
def augment(x, y):
    x = tf.keras.preprocessing.image.random_shift(x, 0.1, 0.1)
    x = tf.keras.preprocessing.image.random_rotation(
    x, 10, row_axis=1, col_axis=2, channel_axis=0, fill_mode='nearest', cval=0.0,
    interpolation_order=1)
    return x, y

X = np.random.random(size=(256, 48, 48, 1))
y = np.random.randint(0, 7, size=(256,))
dataset = tf.data.Dataset.from_tensor_slices((X, y))
dataset = dataset.map(augment)
dataset = dataset.batch(16, drop_remainder=False)
dataset = dataset.prefetch(buffer_size=1)

Answer 1

在运行您的代码时，我得到了以下错误：AttributeError: 'Tensor' object has no attribute 'ndim'。使用tf.data.Dataset运行augment函数似乎是不可能的，因为它不能处理张量。一种解决方法是将增强函数包装在tf.py_function中

import tensorflow as tf
import numpy as np

def augment(x, y):
    x = x.numpy()
    x = tf.keras.preprocessing.image.random_shift(x, 0.1, 0.1)
    x = tf.keras.preprocessing.image.random_rotation(
    x, 10, row_axis=1, col_axis=2, channel_axis=0, fill_mode='nearest', cval=0.0,
    interpolation_order=1)
    return x, y

X = np.random.random(size=(256, 48, 48, 1))
y = np.random.randint(0, 7, size=(256,))

dataset = tf.data.Dataset.from_tensor_slices((X, y))
dataset = dataset.map(
    lambda x, y: tf.py_function(
        func=augment,
        inp=[x, y],
        Tout=[tf.float32, tf.int64]))
dataset = dataset.batch(16, drop_remainder=False)
dataset = dataset.prefetch(buffer_size=1)

上面的代码应该运行时没有任何错误。如果您经常需要用tf.py_function包装您的函数，那么编写一个装饰器会很方便(也很干净)。如下所示：

import tensorflow as tf
import numpy as np

def map_decorator(func):
    def wrapper(*args):
        return tf.py_function(
            func=func,
            inp=[*args],
            Tout=[a.dtype for a in args])
    return wrapper

@map_decorator
def augment(x, y):
    x = x.numpy()
    x = tf.keras.preprocessing.image.random_shift(x, 0.1, 0.1)
    x = tf.keras.preprocessing.image.random_rotation(
    x, 10, row_axis=1, col_axis=2, channel_axis=0, fill_mode='nearest', cval=0.0,
    interpolation_order=1)
    return x, y

X = np.random.random(size=(256, 48, 48, 1))
y = np.random.randint(0, 7, size=(256,))

dataset = tf.data.Dataset.from_tensor_slices((X, y))
dataset = dataset.map(augment)
dataset = dataset.batch(16, drop_remainder=False)
dataset = dataset.prefetch(buffer_size=1)

希望它能帮上忙！