What does TensorFlow's `conv2d_transpose()` operation do?

Question

The documentation for the conv2d_transpose() operation does not clearly explain what it does:

The transpose of conv2d.

This opera

Answer 1

One application for conv2d_transpose is upscaling, here is an example that explains how it works:

a = np.array([[0, 0, 1.5],
              [0, 1, 0],
              [0, 0, 0]]).reshape(1,3,3,1)

filt = np.array([[1, 2],
                 [3, 4.0]]).reshape(2,2,1,1)

b = tf.nn.conv2d_transpose(a,
                           filt,
                           output_shape=[1,6,6,1],
                           strides=[1,2,2,1],
                           padding='SAME')

print(tf.squeeze(b))

tf.Tensor(
[[0.  0.  0.  0.  1.5 3. ]
 [0.  0.  0.  0.  4.5 6. ]
 [0.  0.  1.  2.  0.  0. ]
 [0.  0.  3.  4.  0.  0. ]
 [0.  0.  0.  0.  0.  0. ]
 [0.  0.  0.  0.  0.  0. ]], shape=(6, 6), dtype=float64)