Michal Januszewski

Thanks for the pull request! Have you considered integrating this directly into run_inference.py? (selectable via a flag). This would help reduce code duplication.

To utilize multiple GPUs during inference, generating independent segmentation for partially overlapping subvolumes is the typical thing to do. These need to be reconciled and assembled into a global segmentation....

We unfortunately don't have this functionality in the main ffn repo, but I'm aware of at least one third party solution (https://github.com/Hanyu-Li/klab_utils/tree/master/klab_utils/ffn/reconciliation) which you might be able to use. IIUC,...

Could you please squash all the commits within the PR into a single one, and also run pyink on the code, using the settings from https://github.com/google/ffn/blob/master/ffn/pyproject.toml ?

Thanks! Unfortunately, it looks like it formatted everything with 4 spaces. One problem might have been that our pyproject.toml got accidentally moved to an incorrect location, so maybe that's why...

> thanks for the detailed inspection! Thanks for your patience with this. We're very close to having this completed now.

Could you please try running https://github.com/google/pytype on this code? We're getting a bunch of failures caused by the new annotations.

Here are the specific failures that are triggered: ``` File "third_party/py/ffn/utils/proofreading.py", line 65, in __init__: Built-in function len was called with the wrong arguments [wrong-arg-types] Expected: (obj: Sized) Actually passed:...

Were you able to get type checking to work on your end? I'm still seeing some failures with the latest change: ``` File "third_party/py/ffn/utils/proofreading.py", line 115, in update_segments: unsupported operand...

Thanks for the last round of changes! The latest version passes all our internal type checks, but I noticed that the annotation for the object list was actually not completely...