tf.random_crop()

tf.random_crop(value, size, seed=None, name=None)

Randomly crops a tensor to a given size.

Slices a shape size portion out of value at a uniformly chosen offset. Requires value.shape >= size.

If a dimension should not be cropped, pass the full size of that dimension. For example, RGB images can be cropped with size = [crop_height, crop_width, 3].

Args:

• value: Input tensor to crop.
• size: 1-D tensor with size the rank of value.
• seed: Python integer. Used to create a random seed. See set_random_seed for behavior.
• name: A name for this operation (optional).

Returns:

A cropped tensor of the same rank as value and shape size.