tf.keras.ops.fold

Combines an array of sliding local blocks into a large containing tensor (reverses unfold).

tf.keras.ops.fold(
    x, output_size, kernel_size, dilation=1, padding=0, stride=1
)

This operation is known as col2im when used with convolution. It takes a 3-D tensor of flattened patches and reconstructs a 4-D image tensor by summing overlapping patches.

x A 3-D tensor of shape (N, C * kH * kW, L) where L is the total number of blocks.
output_size int or tuple of two ints (oH, oW), the spatial shape of the output tensor.
kernel_size int or tuple of two ints, the size of the sliding window (kH, kW). If a single int is given, it is used for both dimensions.
dilation int or tuple of two ints, the spacing between kernel points. Default: 1.
padding int or tuple of two ints, the amount of zero-padding that was applied to the input of unfold. Default: 0.
stride int or tuple of two ints, the step size of the sliding window. Default: 1.

A 4-D tensor of shape (N, C, oH, oW).

Example:

x = keras.ops.ones((1, 2, 4, 4))
patches = keras.ops.unfold(x, kernel_size=2, stride=2)
patches.shape
(1, 8, 4)
y = keras.ops.fold(patches, output_size=(4, 4),
    kernel_size=2, stride=2)
y.shape
(1, 2, 4, 4)