tf.keras.preprocessing.image.random_rotation(
    x, rg, row_axis=1, col_axis=2, channel_axis=0, fill_mode='nearest', cval=0.0,
    interpolation_order=1
)

x: Input tensor. Must be 3D.
rg: Rotation range, in degrees.
row_axis: Index of axis for rows in the input tensor.
col_axis: Index of axis for columns in the input tensor.
channel_axis: Index of axis for channels in the input tensor.
fill_mode: Points outside the boundaries of the input
    are filled according to the given mode
    (one of `{'constant', 'nearest', 'reflect', 'wrap'}`).
cval: Value used for points outside the boundaries
    of the input if `mode='constant'`.
interpolation_order: int, order of spline interpolation.
    see `ndimage.interpolation.affine_transform`

Rotated Numpy image tensor.