docs: update recommended strategy for models with input and output im…

…g and msk) (#2293)

* docs: update recommended strategy for models with input and output (img and msk)

Updates in the custom raster dataset tutorial and the actual file documentation. The previous recommended approach (overriding `__get_item__`) is outdated.

Refs: #2292 (reply in thread)

* fix: grammar and formatting

Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>

* fix: grammar

Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>

---------

Co-authored-by: Mathias Baumgartinger <hiasbaum@gmail.com>
Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>

docs/tutorials/custom_raster_dataset.ipynb

@@ -332,6 +332,10 @@
     "\n",
     "If your data only contains model inputs (such as images), use `is_image = True`. If your data only contains ground truth model outputs (such as segmentation masks), use `is_image = False` instead.\n",
     "\n",
+    "Consequently, the sample returned by the dataset/data loader will use the \"image\" key if *is_image* is True, otherwise it will use the \"mask\" key.\n",
+    "\n",
+    "For datasets with both model inputs and outputs, the recommended approach is to use 2 `RasterDataset` instances and combine them using an `IntersectionDataset`. See L7 Irish, L8 Biome, and I/O Bench for examples of this in `torchgeo/datasets`.\n",
+    "\n",
     "### `dtype`\n",
     "\n",
     "Defaults to float32 for `is_image == True` and long for `is_image == False`. This is what you want for 99% of datasets, but can be overridden for tasks like pixel-wise regression (where the target mask should be float32).\n",

torchgeo/datasets/geo.py

@@ -361,8 +361,8 @@ class RasterDataset(GeoDataset):
     #: The sample returned by the dataset/data loader will use the "image" key if
     #: *is_image* is True, otherwise it will use the "mask" key.
     #:
-    #: For datasets with both model inputs and outputs, a custom
-    #: :func:`~RasterDataset.__getitem__` method must be implemented.
+    #: For datasets with both model inputs and outputs, the recommended approach is
+    #: to use 2 `RasterDataset` instances and combine them using an `IntersectionDataset`.
     is_image = True
 
     #: True if data is stored in a separate file for each band, else False.