Python 随机翻转numpy阵列中每个图像的最有效方法

例如，我有一个numpy数组

a

，其形状

1000*3*256*256

。 换句话说，

a

是1000个图像的数组，每个图像的大小是

3*256*256

。 我想随机翻转每个图像，所以我的问题是如何有效地做到这一点？谢谢

基础：

array[slice（a，b，c）]

相当于

array[a:b:c]

，要反转（“翻转”）数组，请使用

slice（None，None，-1）

，这与

array[：-1]

相同

因此，让我们为每个图像构建随机翻转：

>>> import random
>> flips = [(slice(None, None, None),
...          slice(None, None, random.choice([-1, None])),
...          slice(None, None, random.choice([-1, None])))
...          for _ in xrange(a.shape[0])]

>>> flips[0]
(slice(None, None, None), slice(None, None, -1), slice(None, None, None))
>>> a[0]
array([[[ 0,  1,  2,  3,  4],
        [ 5,  6,  7,  8,  9],
        [10, 11, 12, 13, 14],
        [15, 16, 17, 18, 19],
        [20, 21, 22, 23, 24]],

       [[25, 26, 27, 28, 29],
        [30, 31, 32, 33, 34],
        [35, 36, 37, 38, 39],
        [40, 41, 42, 43, 44],
        [45, 46, 47, 48, 49]]])
>>> a[0][flips[0]]
array([[[20, 21, 22, 23, 24],
        [15, 16, 17, 18, 19],
        [10, 11, 12, 13, 14],
        [ 5,  6,  7,  8,  9],
        [ 0,  1,  2,  3,  4]],

       [[45, 46, 47, 48, 49],
        [40, 41, 42, 43, 44],
        [35, 36, 37, 38, 39],
        [30, 31, 32, 33, 34],
        [25, 26, 27, 28, 29]]])

>>> random_flipped = np.array([img[flip] for img, flip in zip(a, flips)])

第一个切片用于通道，第二个切片用于Y轴，第三个切片用于X轴。让我们构建一些测试数据：

>>> import numpy as np
>>> a = np.array(range(3*2*5*5)).reshape(3,2,5,5)

我们可以将每个随机翻转分别应用于每个图像：

>>> import random
>> flips = [(slice(None, None, None),
...          slice(None, None, random.choice([-1, None])),
...          slice(None, None, random.choice([-1, None])))
...          for _ in xrange(a.shape[0])]

>>> flips[0]
(slice(None, None, None), slice(None, None, -1), slice(None, None, None))
>>> a[0]
array([[[ 0,  1,  2,  3,  4],
        [ 5,  6,  7,  8,  9],
        [10, 11, 12, 13, 14],
        [15, 16, 17, 18, 19],
        [20, 21, 22, 23, 24]],

       [[25, 26, 27, 28, 29],
        [30, 31, 32, 33, 34],
        [35, 36, 37, 38, 39],
        [40, 41, 42, 43, 44],
        [45, 46, 47, 48, 49]]])
>>> a[0][flips[0]]
array([[[20, 21, 22, 23, 24],
        [15, 16, 17, 18, 19],
        [10, 11, 12, 13, 14],
        [ 5,  6,  7,  8,  9],
        [ 0,  1,  2,  3,  4]],

       [[45, 46, 47, 48, 49],
        [40, 41, 42, 43, 44],
        [35, 36, 37, 38, 39],
        [30, 31, 32, 33, 34],
        [25, 26, 27, 28, 29]]])

>>> random_flipped = np.array([img[flip] for img, flip in zip(a, flips)])

如您所见，

垂直翻转[0]

垂直翻转图像。现在，为每个图像执行此操作非常简单：

>>> import random
>> flips = [(slice(None, None, None),
...          slice(None, None, random.choice([-1, None])),
...          slice(None, None, random.choice([-1, None])))
...          for _ in xrange(a.shape[0])]

>>> flips[0]
(slice(None, None, None), slice(None, None, -1), slice(None, None, None))
>>> a[0]
array([[[ 0,  1,  2,  3,  4],
        [ 5,  6,  7,  8,  9],
        [10, 11, 12, 13, 14],
        [15, 16, 17, 18, 19],
        [20, 21, 22, 23, 24]],

       [[25, 26, 27, 28, 29],
        [30, 31, 32, 33, 34],
        [35, 36, 37, 38, 39],
        [40, 41, 42, 43, 44],
        [45, 46, 47, 48, 49]]])
>>> a[0][flips[0]]
array([[[20, 21, 22, 23, 24],
        [15, 16, 17, 18, 19],
        [10, 11, 12, 13, 14],
        [ 5,  6,  7,  8,  9],
        [ 0,  1,  2,  3,  4]],

       [[45, 46, 47, 48, 49],
        [40, 41, 42, 43, 44],
        [35, 36, 37, 38, 39],
        [30, 31, 32, 33, 34],
        [25, 26, 27, 28, 29]]])

>>> random_flipped = np.array([img[flip] for img, flip in zip(a, flips)])

---

如果你都在做，怎么可能是随机的呢？你真的写过代码吗？定义“随机翻转每个图像”，实际上这种操作在深度学习文献中称为数据增强。我不想使用for循环，因为它可能非常慢。。。我会尝试@BlackBear的方法，谢谢，我的意思是对每一张图片随机翻转。谢谢，这正是我想要的。