System Info

• transformers version: 4.25.1
• Platform: macOS-11.6.8-x86_64-i386-64bit
• Python version: 3.10.8
• Huggingface_hub version: 0.11.1
• PyTorch version (GPU?): not installed (NA)
• Tensorflow version (GPU?): not installed (NA)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Using GPU in script?: no
• Using distributed or parallel set-up in script?: no

Who can help?

@amyeroberts @NielsRogge

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

Steps to reproduce the behavior:

1. Create and activate clean Python environment

   python3 -m venv venv
   source venv/bin/activate

2. Install transformers and its direct dependencies

   pip install transformers

3. Attempt to import transformers.image_transforms or one of its publicly-documented members

   python -c 'from transformers.image_transforms import center_crop'

4. Encounter a NameError

   None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
   Traceback (most recent call last):
     File "<string>", line 1, in <module>
     File "/Users/miliu/venv/lib/python3.10/site-packages/transformers/image_transforms.py", line 51, in <module>
       def to_channel_dimension_format(image: np.ndarray, channel_dim: Union[ChannelDimension, str]) -> np.ndarray:
   NameError: name 'ChannelDimension' is not defined
   

Expected behavior

Rather than a NameError on import (caused by to_channel_dimension_format()'s signature type annotation containing ChannelDimension, which is conditionally imported only when PIL is available), I would've expected something like transformers.rescale's user experience, where it helpfully recommends installing Pillow when one attempts to use it:

>>> from transformers import rescale
None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
>>> rescale()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/miliu/venv/lib/python3.10/site-packages/transformers/utils/dummy_vision_objects.py", line 14, in rescale
    requires_backends(rescale, ["vision"])
  File "/Users/miliu/venv/lib/python3.10/site-packages/transformers/utils/import_utils.py", line 997, in requires_backends
    raise ImportError("".join(failed))
ImportError: 
rescale requires the PIL library but it was not found in your environment. You can install it with pip:
`pip install pillow`. Please note that you may need to restart your runtime after installation.