python - 如何在Keras中为 'Tensor'对象赋值？

Question

我想通过以下方式为张量变量赋值。但是，我收到一条错误消息:“‘张量’对象不支持项目分配”。

我正在尝试将这些Python代码转换为Keras中的tensorflow。但是，第二行给出了错误

s1 = tf.zeros([5:256:256:3], tf.float64) 
s1[:,:,:,2] = -1

#depth is in shape [5:256,256,1]
lamda = -(depth/s2[:,:,:,2])

x_c = np.around(lamda * s1[:,:,:,0]/step,decimals=2) 
y_c = np.around(lamda * s1[:,:,:,1]/step,decimals=2) 

请告诉我如何解决这个问题？预先感谢您。

Answer 1

TensorFlow 张量对象不可分配。
This question和 this可能会有帮助。

import tensorflow as tf

s1 = tf.Variable(tf.zeros([5,256,256,3], tf.float64))
s2 = tf.Variable(-tf.ones([5,256,256,3], tf.float64))  

assign_op = tf.assign(s1[:,:,:,2], s2[:,:,:,2])

init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
result = sess.run(assign_op)

print(result)

[[[[ 0.  0. -1.]
   [ 0.  0. -1.]
   [ 0.  0. -1.]
   ...
   [ 0.  0. -1.]
   [ 0.  0. -1.]
   [ 0.  0. -1.]]

  [[ 0.  0. -1.]
   [ 0.  0. -1.]
   [ 0.  0. -1.]
   ...