Question

我m new to tensorflow and python and I created a feed forward neural network with tensorflow that will help me to classify two groups of images. One group represents images of myself and another group represents images of a different person (I know convolutional network is better for this kind of problem but for the sake of learning I approached the FF network). All my images are stored in two separate directories. I尝试加载图像进行训练并将其输入NN。我的图像是272x272px RGB，所以输入层应该有73984个神经元。我无法加载图像并通过网络提供它们。

我尝试使用这种方法：

filename_queue = tf.train.string_input_producer(tf.train.match_filenames_once("images/train/resized/*.jpg"))

reader = tf.WholeFileReader()
filename, content = reader.read(filename_queue)
image = tf.image.decode_jpeg(content, channels=3) 
image = tf.cast(image, tf.float32) 
resized_image = tf.image.resize_images(image, [272, 272])

然后，当我跑：

sess.run([optimizer], feed_dict={x: resized_image, y: 1})

I gen“Feed的值不能是tf.Tensor对象”错误

有更好的方法可以做到这一点，或者我在这里缺少什么？谢谢！

Answer 1

您的resized_image变量是一个张量，因为您已将其初始化为tf.image.resize_images(image, [272, 272]) ...您的Feed必须是一个numpy数组，其形状必须与您拥有的张量x相匹配在代码中定义例如，如果在你的情况下x = tf.placeholder(tf.float32, (None, 272, 272, 3))然后你必须给它形状必须是形状的bacth（bacth_number，272,272,3）

我建议您使用以下代码来阅读图片...以此代码为例

import matplotlib.image as mpimg
image = mpimg.imread(path_to_the_image)
x = tf.placeholder(tf.float32, (None, 272, 272, 3))

并在会话中运行时：

sess.run([optimizer], feed_dict={x: image.reshape((1, image.shape[0], image.shape[1], image.shape[2])), y: 1})

在张量流中加载自定义图像

1 个答案: