Add South America Soybean DataModule (microsoft#1959)

* Add South America Soybean DataModule

* Add train_aug

* Regenerate data

---------

Co-authored-by: Adam J. Stewart <[email protected]>

docs/api/datamodules.rst

@@ -36,6 +36,7 @@ Sentinel
 
 .. autoclass:: Sentinel2CDLDataModule
 .. autoclass:: Sentinel2NCCMDataModule
+.. autoclass:: Sentinel2SouthAmericaSoybeanDataModule
 
 Non-geospatial DataModules
 --------------------------

tests/conf/sentinel2_south_america_soybean.yaml

@@ -0,0 +1,17 @@
+model:
+ class_path: SemanticSegmentationTask
+ init_args:
+   loss: "ce"
+   model: "deeplabv3+"
+   backbone: "resnet18"
+   in_channels: 13
+   num_classes: 2
+   num_filters: 1
+data:
+ class_path: Sentinel2SouthAmericaSoybeanDataModule
+ init_args:
+   batch_size: 2
+   patch_size: 16
+ dict_kwargs:
+   south_america_soybean_paths: "tests/data/south_america_soybean"
+   sentinel2_paths: "tests/data/sentinel2"

tests/data/south_america_soybean/data.py

@@ -11,7 +11,7 @@
 from rasterio.crs import CRS
 from rasterio.transform import Affine
 
-SIZE = 32
+SIZE = 128
 
 
 np.random.seed(0)
@@ -24,15 +24,8 @@ def create_file(path: str, dtype: str):
         "driver": "GTiff",
         "dtype": dtype,
         "count": 1,
-        "crs": CRS.from_epsg(4326),
-        "transform": Affine(
-            0.0002499999999999943131,
-            0.0,
-            -82.0005000000000024,
-            0.0,
-            -0.0002499999999999943131,
-            0.0005000000000000,
-        ),
+        "crs": CRS.from_epsg(32616),
+        "transform": Affine(10, 0.0, 399960.0, 0.0, -10, 4500000.0),
         "height": SIZE,
         "width": SIZE,
         "compress": "lzw",

tests/trainers/test_segmentation.py

@@ -75,6 +75,7 @@ class TestSemanticSegmentationTask:
             "sen12ms_s2_reduced",
             "sentinel2_cdl",
             "sentinel2_nccm",
+            "sentinel2_south_america_soybean",
             "spacenet1",
             "ssl4eo_l_benchmark_cdl",
             "ssl4eo_l_benchmark_nlcd",

torchgeo/datamodules/__init__.py

@@ -31,6 +31,7 @@
 from .sen12ms import SEN12MSDataModule
 from .sentinel2_cdl import Sentinel2CDLDataModule
 from .sentinel2_nccm import Sentinel2NCCMDataModule
+from .sentinel2_south_america_soybean import Sentinel2SouthAmericaSoybeanDataModule
 from .skippd import SKIPPDDataModule
 from .so2sat import So2SatDataModule
 from .spacenet import SpaceNet1DataModule
@@ -53,6 +54,7 @@
     "NAIPChesapeakeDataModule",
     "Sentinel2CDLDataModule",
     "Sentinel2NCCMDataModule",
+    "Sentinel2SouthAmericaSoybeanDataModule",
     # NonGeoDataset
     "BigEarthNetDataModule",
     "ChaBuDDataModule",

torchgeo/datamodules/sentinel2_south_america_soybean.py

@@ -0,0 +1,123 @@
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License.
+
+
+"""South America Soybean datamodule."""
+
+
+from typing import Any, Optional, Union
+
+import kornia.augmentation as K
+import torch
+from kornia.constants import DataKey, Resample
+from matplotlib.figure import Figure
+
+from ..datasets import Sentinel2, SouthAmericaSoybean, random_grid_cell_assignment
+from ..samplers import GridGeoSampler, RandomBatchGeoSampler
+from ..samplers.utils import _to_tuple
+from ..transforms import AugmentationSequential
+from .geo import GeoDataModule
+
+
+class Sentinel2SouthAmericaSoybeanDataModule(GeoDataModule):
+    """LightningDataModule for SouthAmericaSoybean and Sentinel2 datasets.
+
+    .. versionadded:: 0.6
+    """
+
+    def __init__(
+        self,
+        batch_size: int = 64,
+        patch_size: Union[int, tuple[int, int]] = 64,
+        length: Optional[int] = None,
+        num_workers: int = 0,
+        **kwargs: Any,
+    ) -> None:
+        """Initialize a new Sentinel2SouthAmericaSoybeanDataModule instance.
+
+        Args:
+            batch_size: Size of each mini-batch.
+            patch_size: Size of each patch, either ``size`` or ``(height, width)``.
+            length: Length of each training epoch.
+            num_workers: Number of workers for parallel data loading.
+            **kwargs: Additional keyword arguments passed to
+                :class:`~torchgeo.datasets.SouthAmericaSoybean`
+                (prefix keys with ``south_america_soybean_``) and
+                :class:`~torchgeo.datasets.Sentinel2`
+                (prefix keys with ``sentinel2_``).
+        """
+        self.south_america_soybean_kwargs = {}
+        self.sentinel2_kwargs = {}
+        for key, val in kwargs.items():
+            if key.startswith("south_america_soybean_"):
+                self.south_america_soybean_kwargs[key[22:]] = val
+            elif key.startswith("sentinel2_"):
+                self.sentinel2_kwargs[key[10:]] = val
+
+        super().__init__(
+            SouthAmericaSoybean,
+            batch_size=batch_size,
+            patch_size=patch_size,
+            length=length,
+            num_workers=num_workers,
+            **kwargs,
+        )
+
+        self.train_aug = AugmentationSequential(
+            K.Normalize(mean=self.mean, std=self.std),
+            K.RandomResizedCrop(_to_tuple(self.patch_size), scale=(0.6, 1.0)),
+            K.RandomVerticalFlip(p=0.5),
+            K.RandomHorizontalFlip(p=0.5),
+            data_keys=["image", "mask"],
+            extra_args={
+                DataKey.MASK: {"resample": Resample.NEAREST, "align_corners": None}
+            },
+        )
+
+        self.aug = AugmentationSequential(
+            K.Normalize(mean=self.mean, std=self.std), data_keys=["image", "mask"]
+        )
+
+    def setup(self, stage: str) -> None:
+        """Set up datasets and samplers.
+
+        Args:
+            stage: Either 'fit', 'validate', 'test', or 'predict'.
+        """
+        self.sentinel2 = Sentinel2(**self.sentinel2_kwargs)
+        self.south_america_soybean = SouthAmericaSoybean(
+            **self.south_america_soybean_kwargs
+        )
+        self.dataset = self.sentinel2 & self.south_america_soybean
+
+        generator = torch.Generator().manual_seed(1)
+        (self.train_dataset, self.val_dataset, self.test_dataset) = (
+            random_grid_cell_assignment(
+                self.dataset, [0.8, 0.1, 0.1], grid_size=8, generator=generator
+            )
+        )
+
+        if stage in ["fit"]:
+            self.train_batch_sampler = RandomBatchGeoSampler(
+                self.train_dataset, self.patch_size, self.batch_size, self.length
+            )
+        if stage in ["fit", "validate"]:
+            self.val_sampler = GridGeoSampler(
+                self.val_dataset, self.patch_size, self.patch_size
+            )
+        if stage in ["test"]:
+            self.test_sampler = GridGeoSampler(
+                self.test_dataset, self.patch_size, self.patch_size
+            )
+
+    def plot(self, *args: Any, **kwargs: Any) -> Figure:
+        """Run SouthAmericaSoybean plot method.
+
+        Args:
+            *args: Arguments passed to plot method.
+            **kwargs: Keyword arguments passed to plot method.
+
+        Returns:
+            A matplotlib Figure with the image, ground truth, and predictions.
+        """
+        return self.south_america_soybean.plot(*args, **kwargs)