[BUG]: PILLOW_VERSION

Thank you for helping us improve solaris!

Summary of the bug

There appears to be an issue with PIL version

Steps to reproduce the bug

By following the installation instructions you get the following error:

  conda env create -n solaris -f environment-gpu.yml
  conda activate solaris
  pip install .
  solaris_run_ml --help

...
ImportError: cannot import name 'PILLOW_VERSION' from 'PIL' (anaconda3/envs/solaris/lib/python3.7/site-packages/PIL/__init__.py)

Installing Pillow 6.1 appears to resolve the issue:

pip install Pillow==6.1
solaris_run_ml --help

usage: solaris_run_ml [-h] --config CONFIG

Run a Solaris ML pipeline based on a config YAML

optional arguments:
  -h, --help            show this help message and exit
  --config CONFIG, -c CONFIG
                        Full path to a YAML-formatted config file specifying
                        parameters for model training and/or inference.

Same behavior occurred in master and v0.2.0

[hugovk]

torchvision v0.5.0 has been released with the fix:

1. Require torchvision>=0.5.0
2. If Pillow was temporarily pinned, remove the pin

[nrweir]

Thanks for the heads up @hugovk. In this case I’ll probably let users manage it themselves rather than explicitly pin in the repo since it’ll be fixed in a few days.