Add Digital typhoon dataset #1748
Conversation
github-actions
bot
added
datasets
|
This is really cool! I wonder if there is any generalization between this and the Cyclone dataset |
Stay tuned:) |
github-actions
bot
added
the
documentation
| ) | ||
| ) | ||
|
|
||
| # torchgeo expects a single label |
There was a problem hiding this comment.
We can fix this if needed. I'm trying to add support for multi-label classification in #2219
| # tensor with added channel dimension | ||
| tensor = torch.from_numpy(h5f['Infrared'][:]).unsqueeze(0) | ||
|
|
||
| # follow normalization procedure |
There was a problem hiding this comment.
Wondering if this should be in the datamodule instead. Possibly also true for other stuff.
There was a problem hiding this comment.
I think it has to be part of the dataset because the normalization values can change. In the datamodule it will become quiet messy, and overall I think it's much nicer if normalization happens in the dataset and any augmentations in the datamodule.
There was a problem hiding this comment.
Only complaint is that the user can't perform their own normalization, or if they do it is now duplicated. Definitely something worth discussing more broadly though (not just in your PR, for all datasets).
| name: torch.tensor(feature_df[name].item()).float() | ||
| for name in self.features | ||
| } | ||
| # normalize the targets for regression |
|
Inspired by this rename I |
This PR adds the Digital Typhoon Dataset.
The implementation allows the following features:
TODO:
Sample Image:
