使用tensorflow進行簡單的刪減增強，為張量分配值時出現問題

我想對我的影像進行簡單的裁剪增強，但我對張量不太熟悉，所以我不知道如何做到這一點。

這是代碼：

def augment_cutout(image, label, size=68, n_squares=1):
    h, w, channels = image.shape
    new_image = image
    for _ in range(n_squares):
        y = np.random.randint(h)
        x = np.random.randint(w)
        y1 = np.clip(y - size // 2, 0, h)
        y2 = np.clip(y   size // 2, 0, h)
        x1 = np.clip(x - size // 2, 0, w)
        x2 = np.clip(x   size // 2, 0, w)
        new_image[y1:y2,x1:x2,:] = 0
    return tf.cast(new_image, tf.float32), label

train_dataset = train_dataset.map(map_func = augment_cutout, num_parallel_calls=AUTOTUNE)

上面的代碼導致以下錯誤

<ipython-input-209-308483a55fd4>:59 augment_cutout  *
        new_image = new_image[y1:y2,x1:x2,:].assign(0)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/ops.py:401 __getattr__
        self.__getattribute__(name)

    AttributeError: 'Tensor' object has no attribute 'assign'

有誰知道如何解決這個問題？

編輯：我嘗試過 tensorflow 插件隨機剪切功能，但這也不起作用。

import tensorflow_addons as tfa

def random_cut_out(images, labels):
    return tfa.image.random_cutout(images, (64, 64), constant_values = 1), labels

train_dataset = train_ds.map(map_func = preprocess_img, num_parallel_calls=AUTOTUNE) 
train_dataset = train_dataset.map(map_func = augment, num_parallel_calls=AUTOTUNE)
train_dataset = train_dataset.map(random_cut_out)

我收到以下錯誤：

    ValueError: slice index 3 of dimension 0 out of bounds. for '{{node 

cutout/strided_slice_2}} = StridedSlice[Index=DT_INT32, T=DT_INT32, 

begin_mask=0, ellipsis_mask=0, end_mask=0, new_axis_mask=0, 

shrink_axis_mask=1](cutout/Shape, cutout/strided_slice_2/stack, 

cutout/strided_slice_2/stack_1, cutout/strided_slice_2/stack_2)' with input

 shapes: [3], [1], [1], [1] and with computed input tensors: input[1] = <3>,

 input[2] = <4>, input[3] = <1>.

有關我如何創建資料集的更多資訊：

import tensorflow_datasets as tfds
builder = tfds.ImageFolder('/content/dataset2')
print(builder.info) 
train_ds, val_ds, test_ds = builder.as_dataset(split=['train', 'val', 'test'], shuffle_files=True, as_supervised=True)

def preprocess_img(image, label, img_shape=160):
  image = tf.image.resize(image, [img_shape, img_shape]) # reshape to img_shape
  return tf.cast(image, tf.float32), label # return (float32_image, label) tuple

BATCH_SIZE = 256
AUTOTUNE = tf.data.AUTOTUNE


train_dataset = train_ds.map(map_func = preprocess_img, num_parallel_calls=AUTOTUNE)

train_dataset = train_dataset.shuffle(buffer_size=1000).batch(batch_size=BATCH_SIZE).prefetch(buffer_size=AUTOTUNE)

uj5u.com熱心網友回復：

使用tfa.image.random_cutout應該完全符合您的要求：

import tensorflow as tf
import matplotlib.pyplot as plt
import tensorflow_addons as tfa

def random_cut_out(images, labels):
    return tfa.image.random_cutout(images, (64, 64), constant_values = 1), labels

flowers = tf.keras.utils.get_file(
    'flower_photos',
    'https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz',
    untar=True)

img_gen = tf.keras.preprocessing.image.ImageDataGenerator(rescale=1./255)

ds = tf.data.Dataset.from_generator(
    lambda: img_gen.flow_from_directory(flowers, batch_size=32, shuffle=True),
    output_types=(tf.float32, tf.float32))

ds = ds.map(random_cut_out)
images, _ = next(iter(ds.take(1)))
image = images[0]

plt.imshow(image.numpy())

Found 3670 images belonging to 5 classes.
<matplotlib.image.AxesImage at 0x7f85f5692210>

該tfa.image.random_cutout函式需要一個 shape 的張量(batch_size, height, width, channels)。因此，當您將單個影像輸入到自定義函式中時，您需要添加一個額外的維度：

def random_cut_out(image, label):
    image = tf.expand_dims(image, axis=0)
    return tf.squeeze(tfa.image.random_cutout(image, (64, 64), constant_values = 1), axis=0), label

標籤：Python 张量流 深度学习 张量 数据增强

上一篇：卷積神經網路-1D-特征分類錯誤

下一篇：如何使用tensorflow/keras計算梯度？