module 'PIL.Image' has no attribute 'ANTIALIAS'

[FuFuDav]

Environment information (required)

Linux:

cuda 11.4
pytorch 1.12.1
Python 3.8.18
PIL 10.3.0
tensorboard 2.14.0

I setup the environment by :
$ conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch

Issue description

When using writer.add_figure to add a figure, there is a Error:

AttributeError: module 'PIL.Image' has no attribute 'ANTIALIAS'

The original code line in my project is:
writer.add_figure('indicator/CM', fig)
fig is a matplotlib Figure, where a confusion matrix is drawn.

It seems in the process of add_images of add_figure , it is attempted to get the attribute ANTIALIAS of PIL.Image. BUT in the recent update, it was removed! Since version 10.0.0, ANTIALIAS was already removed, but add_figure method still needs it.

Reproduce the problem

Add a figure to a writer. The figure draws a confusion matrix.
This should reproduce the problem.

[arcra]

Can you add some details about what "writer" you're referring to (what class or implementation you're using)? And share a stack trace?

I'm not sure this is an issue with our code specifically. This might be a pytorch-specific implementation. But those details would help figuring it out.