Remove dataset-specific trainers

conf/task_defaults/eurosat.yaml

@@ -7,7 +7,7 @@ experiment:
     learning_rate_schedule_patience: 6
     weights: "random"
     in_channels: 13
-    num_classes: 10
+    num_classes: 2
   datamodule:
     root_dir: "tests/data/eurosat"
     batch_size: 1

conf/task_defaults/resisc45.yaml

@@ -7,7 +7,7 @@ experiment:
     learning_rate_schedule_patience: 6
     weights: "random"
     in_channels: 3
-    num_classes: 45
+    num_classes: 3
   datamodule:
     root_dir: "tests/data/resisc45"
     batch_size: 1

conf/task_defaults/ucmerced.yaml

@@ -7,7 +7,7 @@ experiment:
     learning_rate: 1e-3
     learning_rate_schedule_patience: 6
     in_channels: 3
-    num_classes: 21
+    num_classes: 2
   datamodule:
     root_dir: "tests/data/ucmerced"
     batch_size: 1

docs/api/datasets.rst

@@ -270,3 +270,4 @@ Collation Functions
 .. autofunction:: stack_samples
 .. autofunction:: concat_samples
 .. autofunction:: merge_samples
+.. autofunction:: unbind_samples

tests/datamodules/test_landcoverai.py

@@ -0,0 +1,27 @@
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License.
+
+import os
+from typing import Any, Dict, cast
+
+import pytest
+from omegaconf import OmegaConf
+
+from torchgeo.datamodules import LandCoverAIDataModule
+
+
+class TestLandCoverAIDataModule:
+    @pytest.fixture(scope="class")
+    def datamodule(self) -> LandCoverAIDataModule:
+        conf = OmegaConf.load(os.path.join("conf", "task_defaults", "landcoverai.yaml"))
+        kwargs = OmegaConf.to_object(conf.experiment.datamodule)
+        kwargs = cast(Dict[str, Any], kwargs)
+
+        datamodule = LandCoverAIDataModule(**kwargs)
+        datamodule.prepare_data()
+        datamodule.setup()
+        return datamodule
+
+    def test_undefined_attribute(self, datamodule: LandCoverAIDataModule) -> None:
+        batch = next(iter(datamodule.train_dataloader()))
+        batch = datamodule.on_after_batch_transfer(batch, 0)

tests/datamodules/test_resisc45.py

@@ -0,0 +1,27 @@
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License.
+
+import os
+from typing import Any, Dict, cast
+
+import pytest
+from omegaconf import OmegaConf
+
+from torchgeo.datamodules import RESISC45DataModule
+
+
+class TestRESISC45DataModule:
+    @pytest.fixture(scope="class")
+    def datamodule(self) -> RESISC45DataModule:
+        conf = OmegaConf.load(os.path.join("conf", "task_defaults", "resisc45.yaml"))
+        kwargs = OmegaConf.to_object(conf.experiment.datamodule)
+        kwargs = cast(Dict[str, Any], kwargs)
+
+        datamodule = RESISC45DataModule(**kwargs)
+        datamodule.prepare_data()
+        datamodule.setup()
+        return datamodule
+
+    def test_undefined_attribute(self, datamodule: RESISC45DataModule) -> None:
+        batch = next(iter(datamodule.train_dataloader()))
+        batch = datamodule.on_after_batch_transfer(batch, 0)

tests/datasets/test_utils.py

@@ -31,6 +31,7 @@
     merge_samples,
     percentile_normalization,
     stack_samples,
+    unbind_samples,
     working_dir,
 )
 
@@ -457,7 +458,7 @@ def samples(self) -> List[Dict[str, Any]]:
             },
         ]
 
-    def test_stack_samples(self, samples: List[Dict[str, Any]]) -> None:
+    def test_stack_unbind_samples(self, samples: List[Dict[str, Any]]) -> None:
         sample = stack_samples(samples)
         assert sample["image"].size() == torch.Size(  # type: ignore[attr-defined]
             [2, 3]
@@ -468,6 +469,13 @@ def test_stack_samples(self, samples: List[Dict[str, Any]]) -> None:
         )
         assert sample["crs"] == [CRS.from_epsg(2000), CRS.from_epsg(2001)]
 
+        new_samples = unbind_samples(sample)
+        for i in range(2):
+            assert torch.allclose(  # type: ignore[attr-defined]
+                samples[i]["image"], new_samples[i]["image"]
+            )
+            assert samples[i]["crs"] == new_samples[i]["crs"]
+
     def test_concat_samples(self, samples: List[Dict[str, Any]]) -> None:
         sample = concat_samples(samples)
         assert sample["image"].size() == torch.Size([6])  # type: ignore[attr-defined]
@@ -500,7 +508,7 @@ def samples(self) -> List[Dict[str, Any]]:
             },
         ]
 
-    def test_stack_samples(self, samples: List[Dict[str, Any]]) -> None:
+    def test_stack_unbind_samples(self, samples: List[Dict[str, Any]]) -> None:
         sample = stack_samples(samples)
         assert sample["image"].size() == torch.Size(  # type: ignore[attr-defined]
             [1, 3]
@@ -515,6 +523,16 @@ def test_stack_samples(self, samples: List[Dict[str, Any]]) -> None:
         assert sample["crs1"] == [CRS.from_epsg(2000)]
         assert sample["crs2"] == [CRS.from_epsg(2001)]
 
+        new_samples = unbind_samples(sample)
+        assert torch.allclose(  # type: ignore[attr-defined]
+            samples[0]["image"], new_samples[0]["image"]
+        )
+        assert samples[0]["crs1"] == new_samples[0]["crs1"]
+        assert torch.allclose(  # type: ignore[attr-defined]
+            samples[1]["mask"], new_samples[0]["mask"]
+        )
+        assert samples[1]["crs2"] == new_samples[0]["crs2"]
+
     def test_concat_samples(self, samples: List[Dict[str, Any]]) -> None:
         sample = concat_samples(samples)
         assert sample["image"].size() == torch.Size([3])  # type: ignore[attr-defined]

tests/trainers/test_classification.py

@@ -50,6 +50,23 @@ def test_trainer(self, name: str, classname: Type[LightningDataModule]) -> None:
         trainer.fit(model=model, datamodule=datamodule)
         trainer.test(model=model, datamodule=datamodule)
 
+    def test_no_logger(self) -> None:
+        conf = OmegaConf.load(os.path.join("conf", "task_defaults", "ucmerced.yaml"))
+        conf_dict = OmegaConf.to_object(conf.experiment)
+        conf_dict = cast(Dict[Any, Dict[Any, Any]], conf_dict)
+
+        # Instantiate datamodule
+        datamodule_kwargs = conf_dict["datamodule"]
+        datamodule = UCMercedDataModule(**datamodule_kwargs)
+
+        # Instantiate model
+        model_kwargs = conf_dict["module"]
+        model = ClassificationTask(**model_kwargs)
+
+        # Instantiate trainer
+        trainer = Trainer(logger=None, fast_dev_run=True, log_every_n_steps=1)
+        trainer.fit(model=model, datamodule=datamodule)
+
     @pytest.fixture
     def model_kwargs(self) -> Dict[Any, Any]:
         return {
@@ -120,6 +137,25 @@ def test_trainer(self, name: str, classname: Type[LightningDataModule]) -> None:
         trainer.fit(model=model, datamodule=datamodule)
         trainer.test(model=model, datamodule=datamodule)
 
+    def test_no_logger(self) -> None:
+        conf = OmegaConf.load(
+            os.path.join("conf", "task_defaults", "bigearthnet_s1.yaml")
+        )
+        conf_dict = OmegaConf.to_object(conf.experiment)
+        conf_dict = cast(Dict[Any, Dict[Any, Any]], conf_dict)
+
+        # Instantiate datamodule
+        datamodule_kwargs = conf_dict["datamodule"]
+        datamodule = BigEarthNetDataModule(**datamodule_kwargs)
+
+        # Instantiate model
+        model_kwargs = conf_dict["module"]
+        model = MultiLabelClassificationTask(**model_kwargs)
+
+        # Instantiate trainer
+        trainer = Trainer(logger=None, fast_dev_run=True, log_every_n_steps=1)
+        trainer.fit(model=model, datamodule=datamodule)
+
     @pytest.fixture
     def model_kwargs(self) -> Dict[Any, Any]:
         return {