如何在张量流中使用张量索引另一个张量

Question

I have a data tensor of dimensios [BXNX 3] , and I have an indices tensor of dimensions [BXM] . I wish to extract a [BXMX 3] tensor from the data tensor using the indices tensor.

I have this code that works :

new_data= []    
for i in range(B):
        new_data.append(tf.gather(data[i], indices[i]))
new_data= tf.stack(new_data) 

However, I am sure it is not the right way to do this. Does anyone know a better way? (I guess I should use tf.gather_nd() somehow but I couldn't figure out how)

I have seen several answers to similar questions here . However I could not find the solution to my problem.

Answer 1

You can use tf.gather_nd() with code like this:

import tensorflow as tf

# B = 3
# N = 4
# M = 2
# [B x N x 3]
data = tf.constant([
    [[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10, 11]],
    [[100, 101, 102], [103, 104, 105], [106, 107, 108], [109, 110, 111]],
    [[200, 201, 202], [203, 204, 205], [206, 207, 208], [209, 210, 211]],
    ])

# [B x M]
indices = tf.constant([
    [0, 2],
    [1, 3],
    [3, 2],
    ])

indices_shape = tf.shape(indices)

indices_help = tf.tile(tf.reshape(tf.range(indices_shape[0]), [indices_shape[0], 1]) ,[1, indices_shape[1]]);
indices_ext = tf.concat([tf.expand_dims(indices_help, 2), tf.expand_dims(indices, 2)], axis = 2)
new_data = tf.gather_nd(data, indices_ext)

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print('data')
    print(sess.run(data))
    print('\nindices')
    print(sess.run(indices))
    print('\nnew_data')
    print(sess.run(new_data))

new_data will be:

[[[  0   1   2]
  [  6   7   8]]

 [[103 104 105]
  [109 110 111]]

 [[209 210 211]
  [206 207 208]]]