Dimitris Mantas

This is a very useful metric for semantic segmentation since it correlates nicely with human perception of IoU. Any news on whether it will be added?

I can do a PR with per-channel min-max scaling to [0,1] as the default augmentation. I'm actually using this type of scaling for my own work but my implementation requires...

Ah, please excuse the confusion; by "per-channel" I meant that if you have an `n`-band raster, each of its channels gets normalised according to its own minimum and maximum values....

I think cubic interpolation is not such a good idea because the corresponding convolution kernel has negative weights, meaning that the output data range is not guaranteed. This can really...

I think this is more or less the one-liner for each method: - Nearest/Mode: Good for masks; not suitable for continuous fields (i.e., images) - Bilinear: Suitable for continuous data;...

By the way, Lanczos is technically "the best" of the bunch...

Given that most metrics of interest are broken (e.g., all of them when ``average="macro"`` and ``ignore_index`` is specified (https://github.com/Lightning-AI/torchmetrics/pull/2443) and``JaccardIndex`` which outputs NaN when ``average==macro`` instead when you try to...

Sure, that makes sense; please excuse the rant haha.

You need to set both ``on_step`` and ``on_epoch`` to get logs only per step or per epoch.

It just hit me that we should be a bit careful with which metrics we add to avoid unnecessary computation; class-wise accuracy and recall are the same thing and so...