pyflink.multimodal.codec.image_to_ndarray

image_to_ndarray(image, mode=None)

Return decoded Image pixels as an ndarray for operator internals.

When mode is None, this returns the stored Image data unchanged, including its canonical 3-D (height, width, channels) shape. Callers may rely on this 3-D HWC shape; single-channel decoded images must not be collapsed to 2-D in the mode=None path. When a target mode is provided, Pillow handles color/channel conversion and the codec applies linear 8-bit/16-bit scaling. Single-channel targets such as L and L16 are returned in Pillow’s 2-D array shape.