Note the available backends in docstring of `keras.preprocessing.image.smart_resize()`

[ILCSFNO]

Doc Issue

The doc of keras.preprocessing.image.smart_resize() shows its description as below:

keras/keras/src/utils/image_utils.py

Lines 357 to 358 in 3fac66f

	backend_module: Backend module to use (if different from the default
	backend).

For the repro below, I know that the default backend is keras.src.backend, but what backend can be served as other Backend module to use?

I tried keras.backend but failed. If there are some other backends available to use, I suggest to note some examples of them in the docstring above.

Thanks a lot!

Repro

import numpy as np
import keras
size = (256, 256)
backend_module = keras.backend
image = np.random.randint(0, 256, size=(size[0], size[1], 3), dtype=np.uint8)
resized_image = keras.preprocessing.image.smart_resize(image, size, interpolation='bicubic', backend_module=backend_module)
print('Original Image Shape:', image.shape)
print('Resized Image Shape:', resized_image.shape)

Output

AttributeError: module 'keras.api.backend' has no attribute 'convert_to_tensor'

[Flakes342]

Good catch @ILCSFNO. The parameter is misleading as currently documented and should be clarified to prevent any confusion. The backend_module parameter only accepts keras.src.backend (which is already the default). There are no other valid options, I also tried tf, jax, etc.

Why keras.backend doesn't work: The function internally calls backend_module.image.resize(), which requires the internal backend structure.

keras/keras/src/utils/image_utils.py

Line 451 in 3fac66f

img = backend_module.image.resize(

The public keras.backend API doesn't have this image submodule. So, I guess that the docstring should be updated to state this clearly.