SSL4EO Landsat Downstream Dataset/module CDL, NLCD #1338
Conversation
github-actions
bot
added
datasets
| """ | ||
| super().__init__(SSL4EOLBenchmark, batch_size, num_workers, **kwargs) | ||
|
|
||
| self.train_aug = AugmentationSequential( |
There was a problem hiding this comment.
If you use the same aug for all splits, you can just set self.aug instead. However, we might want different augs for train to improve performance.
There was a problem hiding this comment.
Yes, I wrote them out because I thought it remains to be discussed whether we have additional training augs.
| ) | ||
| exists.append(bool(glob.glob(mask_pathname, recursive=True))) | ||
| if all(exists): | ||
| return |
There was a problem hiding this comment.
Let's put a blank line between sections
There was a problem hiding this comment.
I meant between the return statement and the comment below
| plt_cmap = ListedColormap( | ||
| np.stack( | ||
| [np.array(val) / 255 for val in self.cmaps[self.mask_product].values()], | ||
| axis=0, | ||
| ) | ||
| ) |
There was a problem hiding this comment.
Forgot to say this during the CDL review, but don't we also need to map this to 134 classes before plotting? Should have asked you to add an example plot for CDL.
There was a problem hiding this comment.
Ah sorry, now I do get what you mean. Yeah you are right.
There was a problem hiding this comment.
In NLCD the cmap is already done with the ordinal values. I can fix it in CDL in another PR.
This PR adds datasets and module for downstream evaluation of SSL methods.
Sensors:
Masks:
This is a
NonGeoDatasetrelying on datasets to be created with the help of #1336.Example CDL TM-TOA: