You must feed a value for placeholder tensor 'input/noise' with d-type float and shape [3,3]

环境：tensorflow1.x python3.6
目的：想要在包含tf.variable-scope() 语句的后边 把一个张量转换为numpy 总是报错‘You must feed a value for placeholder tensor 'input/noise' with dtype float and shape [3,3]’
代码：
import tensorflow as tf
import numpy as np

dim = 3
a = np.array([[1, 2, 3], [111, 222, 333], [0, 4, 0]])

b = np.array([[4, 5, 6], [44, 55, 67]])

b = tf.convert_to_tensor(b, tf.float32, name='bob')

define noise-input structure 确保（迭代时）损失曲线快速下降

with tf.variable_scope('input', reuse=True):
rand = tf.placeholder(shape=(dim, dim), dtype=tf.float32, name='noise')
raw_measure = tf.constant(a, dtype=tf.float32, name='diffraction')
c = raw_measure + rand

转化为numpy数组

with tf.Session() as sess:
c = c.eval(session=sess)
print(a)
print(c)

报错：
InvalidArgumentError (see above for traceback): You must feed a value for placeholder tensor 'input/noise' with dtype float and shape [3,3]

解决思路：
按照要求添加feed_dict 但是仍然报错（如上）
with tf.Session() as sess:
# sess.run(tf.global_variables_initializer())
angry = np.random.uniform(0, noise_level, size=(dim, dim))
loss_measure_flu = sess.run(angry1, feed_dict={rand: angry})
inptfp = out.eval(session=sess)

我想要达到的结果：
在包含tf.variable-scope() 语句的后边 把一个张量转换为numpy 总是报错‘You must feed a value for placeholder tensor 'input/noise' with dtype float and shape [3,3]’