Ashwin Nair

> Which model did you use to produce those masks in the image? U-Net with resnet18 backbone. Think I used the exact same config as the one in [`conf/inria.yaml`](https://github.com/microsoft/torchgeo/blob/main/conf/inria.yaml)

> Are we all on the same page with this, or is there disagreement? Yup. I’ve always seen the scripts that way. Thanks @calebrob6. Your solution looks like it might...

Recently came back to this. I incorporated Caleb's solution and wrote a new [`predict.py`](https://gist.github.com/ashnair1/369842c3f9833a3a49a54c8c284d2013) and it does indeed work for georeferenced images. This did require making `preprocess` a class method...

The next release of `torchmetrics` (v0.10.0) will contain `binary`, `multi-class` and `multi-label` variants of all the classification metrics. An RC has already been released.

Forcing binary datasets to have a background class would be the simpler option. I had trained a model on the Inria dataset in the same way (setting num_classes=2) and it...

I believe I only explicitly imported affine for type annotations (in `load_image` and `load_mask`). The image transform of a rasterio image object is an `Affine` object. Even if it was...

SpaceNet 6 is currently unavailable on [Radiant MLHub](https://mlhub.earth/data/spacenet6). I had raised radiantearth/radiant-mlhub#93 to track this among other issues with the SpaceNet datasets. Will give them a ping.

I did notice those attributes and don't think they can be reused. `bands`, `all_bands` and `rgb_bands` are a list of band ~filenames~ names used when bands are stored separately. Suppose...

I think I understand. Do you mean something like this in the `__init__` method? ``` if self.all_bands and self.bands: band_indexes = [self.all_bands.index(i) + 1 for i in self.bands] assert len(band_indexes)...

So we have a `bands` parameter that determines our bands of interest and `all_bands` which are all the bands in the dataset. If we go back to my previous example...