expand_dims

tf.expand_dims  |  TensorFlow https://tensorflow.google.cn/api_docs/python/tf/expand_dims

tf.expand_dims

 

 

tf.expand_dims(
    input,
    axis=None,
    name=None,
    dim=None
)

Defined in tensorflow/python/ops/array_ops.py.

See the guide: Tensor Transformations > Shapes and Shaping

Inserts a dimension of 1 into a tensor's shape.

Given a tensor input, this operation inserts a dimension of 1 at the dimension index axis of input's shape. The dimension index axis starts at zero; if you specify a negative number for axis it is counted backward from the end.

This operation is useful if you want to add a batch dimension to a single element. For example, if you have a single image of shape [height, width, channels], you can make it a batch of 1 image with expand_dims(image, 0), which will make the shape [1, height, width, channels].

Other examples:

 

# 't' is a tensor of shape [2]
tf.shape(tf.expand_dims(t, 0))  # [1, 2]
tf.shape(tf.expand_dims(t, 1))  # [2, 1]
tf.shape(tf.expand_dims(t, -1))  # [2, 1]

# 't2' is a tensor of shape [2, 3, 5]
tf.shape(tf.expand_dims(t2, 0))  # [1, 2, 3, 5]
tf.shape(tf.expand_dims(t2, 2))  # [2, 3, 1, 5]
tf.shape(tf.expand_dims(t2, 3))  # [2, 3, 5, 1]

This operation requires that:

-1-input.dims() <= dim <= input.dims()

This operation is related to squeeze(), which removes dimensions of size 1.

Args:

• input: A Tensor.
• axis: 0-D (scalar). Specifies the dimension index at which to expand the shape of input. Must be in the range [-rank(input) - 1, rank(input)].
• name: The name of the output Tensor.
• dim: 0-D (scalar). Equivalent to axis, to be deprecated.

Returns:

A Tensor with the same data as input, but its shape has an additional dimension of size 1 added.

Raises:

• ValueError: if both dim and axis are specified.