Some pretrained torchvision is only available in torchvision > 0.10.0

[da-liii]

>>> from torchvision.models import efficientnet_b0
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ImportError: cannot import name 'efficientnet_b0' from 'torchvision.models' (/Users/da/.pyenv/versions/rikai/lib/python3.8/site-packages/torchvision/models/__init__.py)
>>> import torchvision
>>> torchvision.__version__
'0.10.0'

From CI, we can know that torchvision 0.12.0 is the workable version.

We have two choices:

• Upgrade torchvision
• Disable the code via torchvision.__version__, throw a RuntimeError like you should use torchvision vx.y.z would be friendly to user.

[eddyxu]

i think it might be easy just bump pytorch and torchvision version

[da-liii]

Python 3.8.10 (default, May 19 2021, 11:01:55)
[Clang 10.0.0 ] :: Anaconda, Inc. on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from torchvision.models import convnext_base
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ImportError: cannot import name 'convnext_base' from 'torchvision.models' (/Users/da/.pyenv/versions/rikai-yolov5/lib/python3.8/site-packages/torchvision/models/__init__.py)
>>>

from torchvision.models import convnext_base does not work for torchvision 0.11.3, but works fine for torchvision 0.12.0.