Question

我之前曾问过这个问题，但经过对问题的一些调查后，我可能刚刚走错了路，试图实现目标。

我想也许这可能是一条更好的尝试之路。但我无法弄清楚的部分是我应该在切片操作上为size参数添加的内容。从根本上说，我想要实现的是能够动态决定如何裁剪图像然后裁剪它然后继续我的计算图中的裁剪图像。如果这似乎是一种效率低下的方法，请随意提供替代方案。

import numpy as np
import tensorflow as tf

img1 = np.random.random([400, 600, 3])
img2 = np.random.random([400, 600, 3])
img3 = np.random.random([400, 600, 3])

images = [img1, img2, img3]

img1_crop = [100, 100, 100, 100]
img2_crop = [200, 150, 100, 100]
img3_crop = [150, 200, 100, 100]

crop_values = [img1_crop, img2_crop, img3_crop]

x = tf.placeholder(tf.float32, shape=[None, 400, 600, 3])
i = tf.placeholder(tf.int32, shape=[None, 4])
y = tf.slice(x, i, size="Not sure what to put here")

# initialize
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)

# run
result = sess.run(y, feed_dict={x: images, i: crop_values})
print(result)

Answer 1

我建议使用tf.image.extract_glimpse，而不是使用tf.slice（不允许您对批处理进行操作）。这是一个批量操作的玩具样本程序：

import tensorflow as tf
import numpy as np

NUM_IMAGES = 2
NUM_CHANNELS = 1
CROP_SIZE = [3, 4]
IMG_HEIGHT=10
IMG_WIDTH=10

# Fake input data, but ordered so we can look at the printed values and
# map them back. The values of the first image are:
# 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],      
#        [50, 51, 52, 53, 54, 55, 56, 57, 58, 59],      
#        [60, 61, 62, 63, 64, 65, 66, 67, 68, 69],      
#        [70, 71, 72, 73, 74, 75, 76, 77, 78, 79],      
#        [80, 81, 82, 83, 84, 85, 86, 87, 88, 89],      
#        [90, 91, 92, 93, 94, 95, 96, 97, 98, 99]])      
image1 = np.reshape(
    np.array(xrange(NUM_IMAGES * IMG_HEIGHT * IMG_WIDTH * NUM_CHANNELS)),
    [NUM_IMAGES, IMG_HEIGHT, IMG_WIDTH, NUM_CHANNELS])

# We use normalized=False to use pixel indexing.
# normalized=True means centers are specified between [0,1).
image1_center = [0, 0]  # The center of the crop is ~ the center of the image.
image2_center = [3, 5]  # Offset down & right in the image.

img = tf.placeholder(tf.float32, shape=[NUM_IMAGES, IMG_HEIGHT, IMG_WIDTH, NUM_CHANNELS], name="img")
size = tf.placeholder(tf.int32, shape=[2], name="crop_size")
centers = tf.placeholder(tf.float32, shape=[NUM_IMAGES, 2], name="centers")
output = tf.image.extract_glimpse(img, size, centers, normalized=False)

sess = tf.Session()
feed_dict = {
  img: image1,
  size: CROP_SIZE,
  centers: [image1_center, image2_center],
}
print sess.run(output, feed_dict=feed_dict)

如果您想提取多种尺寸（甚至每张图片多次瞥见），请查看tf.image.crop_and_resize。

文档：https://www.tensorflow.org/api_docs/python/image/cropping#extract_glimpse

Tensorflow - 动态切片图像

1 个答案: