抽样调查

Question

我是每个图像的采样点使用以下功能。如果tf.range为None，则batch_size将出现错误。如何在tensorflow中取样？

def sampling(binary_selection,num_points, points):
  """
      binary_selection: tensor of size (batch_size, points) 
          with values 1.0 or 0.0. Indicating positive and negative points. 
          We want to sample num_points from positive points of each image
      points: tensor of size (batch_size, num_points_in_image)
      num_points: number of points to sample for each image
  """
  batch_size = points.get_shape()[0]
  indices = tf.multinomial((tf.log(binary_selection)), num_points)
  indices = tf.cast(tf.expand_dims(indices, axis=2), tf.int32)
  batch_seq = tf.expand_dims(tf.range(batch_size), axis=1) 
  im_indices = tf.expand_dims(tf.tile(batch_seq, [1, num_points]), axis=2) 
  indices = tf.concat([im_indices, indices], axis=2)
  return tf.gather_nd(points, indices)

我得到以下错误

_dimension_tensor_conversion_function raise ValueError("Cannot convert an unknown Dimension to a Tensor: %s" % d) ValueError: Cannot convert an unknown Dimension to a Tensor: ?

在测试和培训期间，我将拥有一个整数的batch_size，但是当我初始化时，我希望没有一个作为输入，这样就可以在测试和培训期间改变批处理大小。

Answer 1

您需要向batch_size提供一个值。

它需要初始化。

目前，它没有任何价值。