Remove Kornia master install inside Github Actions (#203)

* remove kornia master install in tests action

* removed unused ignore comments

* removed unused ignore comments x2

.github/workflows/tests.yaml

@@ -19,8 +19,6 @@ jobs:
         python-version: 3.9
     - name: Install pip dependencies
       run: |
-        # TODO: remove after kornia's next release (>0.5.11)
-        pip install git+https://github.com/kornia/kornia@master # required for mypy fixes
         pip install cython numpy  # needed for pycocotools
         pip install .[datasets,tests,train]
     - name: Run mypy checks

benchmark.py

@@ -165,7 +165,7 @@ def main(args: argparse.Namespace) -> None:
         if isinstance(sampler, RandomBatchGeoSampler):
             dataloader = DataLoader(
                 dataset,
-                batch_sampler=sampler,  # type: ignore[arg-type]
+                batch_sampler=sampler,
                 num_workers=args.num_workers,
             )
         else:

tests/samplers/test_batch.py

@@ -55,9 +55,7 @@ class TestBatchGeoSampler:
     @pytest.mark.parametrize("num_workers", [0, 1, 2])
     def test_dataloader(self, sampler: CustomBatchGeoSampler, num_workers: int) -> None:
         ds = CustomGeoDataset()
-        dl = DataLoader(
-            ds, batch_sampler=sampler, num_workers=num_workers  # type: ignore[arg-type]
-        )
+        dl = DataLoader(ds, batch_sampler=sampler, num_workers=num_workers)
         for _ in dl:
             continue
 
@@ -95,8 +93,6 @@ class TestRandomBatchGeoSampler:
     @pytest.mark.parametrize("num_workers", [0, 1, 2])
     def test_dataloader(self, sampler: RandomBatchGeoSampler, num_workers: int) -> None:
         ds = CustomGeoDataset()
-        dl = DataLoader(
-            ds, batch_sampler=sampler, num_workers=num_workers  # type: ignore[arg-type]
-        )
+        dl = DataLoader(ds, batch_sampler=sampler, num_workers=num_workers)
         for _ in dl:
             continue

tests/samplers/test_single.py

@@ -54,9 +54,7 @@ class TestGeoSampler:
     @pytest.mark.parametrize("num_workers", [0, 1, 2])
     def test_dataloader(self, sampler: CustomGeoSampler, num_workers: int) -> None:
         ds = CustomGeoDataset()
-        dl = DataLoader(
-            ds, sampler=sampler, num_workers=num_workers  # type: ignore[arg-type]
-        )
+        dl = DataLoader(ds, sampler=sampler, num_workers=num_workers)
         for _ in dl:
             continue
 
@@ -93,9 +91,7 @@ class TestRandomGeoSampler:
     @pytest.mark.parametrize("num_workers", [0, 1, 2])
     def test_dataloader(self, sampler: RandomGeoSampler, num_workers: int) -> None:
         ds = CustomGeoDataset()
-        dl = DataLoader(
-            ds, sampler=sampler, num_workers=num_workers  # type: ignore[arg-type]
-        )
+        dl = DataLoader(ds, sampler=sampler, num_workers=num_workers)
         for _ in dl:
             continue
 
@@ -135,9 +131,7 @@ class TestGridGeoSampler:
     @pytest.mark.parametrize("num_workers", [0, 1, 2])
     def test_dataloader(self, sampler: GridGeoSampler, num_workers: int) -> None:
         ds = CustomGeoDataset()
-        dl = DataLoader(
-            ds, sampler=sampler, num_workers=num_workers  # type: ignore[arg-type]
-        )
+        dl = DataLoader(ds, sampler=sampler, num_workers=num_workers)
         for _ in dl:
             continue
 

torchgeo/datasets/utils.py

@@ -413,11 +413,11 @@ def dataset_split(
         a list of the subset datasets. Either [train, val] or [train, val, test]
     """
     if test_pct is None:
-        val_length = int(len(dataset) * val_pct)  # type: ignore[arg-type]
-        train_length = len(dataset) - val_length  # type: ignore[arg-type]
+        val_length = int(len(dataset) * val_pct)
+        train_length = len(dataset) - val_length
         return random_split(dataset, [train_length, val_length])
     else:
-        val_length = int(len(dataset) * val_pct)  # type: ignore[arg-type]
-        test_length = int(len(dataset) * test_pct)  # type: ignore[arg-type]
-        train_length = len(dataset) - (val_length + test_length)  # type: ignore[arg-type] # noqa: E501
+        val_length = int(len(dataset) * val_pct)
+        test_length = int(len(dataset) * test_pct)
+        train_length = len(dataset) - (val_length + test_length)
         return random_split(dataset, [train_length, val_length, test_length])

torchgeo/models/changestar.py

@@ -44,7 +44,7 @@ class ChangeMixin(Module):
             num_convs: number of convolution blocks
             scale_factor: number of upsampling factor
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
         layers: List[Module] = [
             nn.modules.Sequential(
                 nn.modules.Conv2d(in_channels, inner_channels, 3, 1, 1),
@@ -130,7 +130,7 @@ class ChangeStar(Module):
                 ``'t2t1'``: concatenate bitemporal features in the order of t2->t1;
                 ``'mean'``: the weighted mean of the output of ``'t1t2'`` and ``'t1t2'``
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
         self.dense_feature_extractor = dense_feature_extractor
         self.seg_classifier = seg_classifier
         self.changemixin = changemixin

torchgeo/models/farseg.py

@@ -64,7 +64,7 @@ class FarSeg(Module):
             classes: number of output segmentation classes
             backbone_pretrained: whether to use pretrained weight for backbone
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
         if backbone in ["resnet18", "resnet34"]:
             max_channels = 512
         elif backbone in ["resnet50", "resnet101"]:
@@ -129,7 +129,7 @@ class _FSRelation(Module):
             in_channels_list: a list of input channels
             out_channels: number of output channels
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
 
         self.scene_encoder = ModuleList(
             [
@@ -160,7 +160,7 @@ class _FSRelation(Module):
                 )
             )
 
-        self.normalizer = Sigmoid()  # type: ignore[no-untyped-call]
+        self.normalizer = Sigmoid()
 
     def forward(self, scene_feature: Tensor, features: List[Tensor]) -> List[Tensor]:
         """Forward pass of the model."""
@@ -202,7 +202,7 @@ class _LightWeightDecoder(Module):
                 levels
             out_feature_output_stride: output stride of output feature maps
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
 
         self.blocks = ModuleList()
         for in_feat_os in in_feature_output_strides:

torchgeo/models/fccd.py

@@ -37,7 +37,7 @@ class ConvBlock(Module):
             dropout: probability for each dropout layer
             pool: max pool last conv layer output if True
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
         layers = []
         for i in range(1, len(channels)):
             layers.extend(
@@ -225,7 +225,7 @@ class FCEF(Module):
             classes: number of output segmentation classes
                 (default=2 for binary segmentation)
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
         self.encoder = Encoder(in_channels * t, pool=True)
         self.decoder = Decoder(classes)
         self.upsample = Upsample()
@@ -267,7 +267,7 @@ class FCSiamConc(Module):
             classes: number of output segmentation classes
                 (default=2 for binary segmentation)
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
         self.encoder = Encoder(in_channels, pool=False)
         self.decoder = ConcatDecoder(t, classes)
         self.upsample = Upsample()
@@ -318,7 +318,7 @@ class FCSiamDiff(nn.modules.Module):
             classes: number of output segmentation classes
                 (default=2 for binary segmentation)
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
         self.encoder = Encoder(in_channels, pool=False)
         self.decoder = Decoder(classes)
         self.upsample = Upsample()

torchgeo/models/fcn.py

@@ -23,7 +23,7 @@ class FCN(Module):
             classes: Number of filters in the final layer
             num_filters: Number of filters in each convolutional layer
         """
-        super(FCN, self).__init__()  # type: ignore[no-untyped-call]
+        super(FCN, self).__init__()
 
         conv1 = nn.modules.Conv2d(
             in_channels, num_filters, kernel_size=3, stride=1, padding=1

torchgeo/models/rcf.py

@@ -43,7 +43,7 @@ class RCF(Module):
             kernel_size: size of the kernel used to compute the RCFs
             bias: bias of the convolutional layer
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
 
         assert features % 2 == 0
 

torchgeo/trainers/byol.py

@@ -56,7 +56,7 @@ class RandomApply(Module):
             augm: augmentation function to apply
             p: probability with which the augmentation function is applied
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
         self.augm = augm
         self.p = p
 
@@ -90,7 +90,7 @@ class SimCLRAugmentation(Module):
         Args:
             image_size: Tuple of integers defining the image size
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
         self.size = image_size
 
         self.augmentation = Sequential(
@@ -128,7 +128,7 @@ class MLP(Module):
             projection_size: size of the output layer
             hidden_size: size of the hidden layer
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
         self.mlp = Sequential(
             Linear(dim, hidden_size),
             BatchNorm1d(hidden_size),  # type: ignore[no-untyped-call]
@@ -174,7 +174,7 @@ class EncoderWrapper(Module):
             hidden_size: size of hidden layer of the projector MLP
             layer: layer from model to project
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
 
         self.model = model
         self.projection_size = projection_size
@@ -275,7 +275,7 @@ class BYOL(Module):
             beta: the speed at which the target encoder is updated using the main
                 encoder
         """
-        super().__init__()  # type: ignore[no-untyped-call]
+        super().__init__()
 
         self.augment: Module
         if augment_fn is None:

torchgeo/trainers/chesapeake.py

@@ -509,7 +509,7 @@ class ChesapeakeCVPRDataModule(LightningDataModule):
         )
         return DataLoader(
             self.train_dataset,
-            batch_sampler=sampler,  # type: ignore[arg-type]
+            batch_sampler=sampler,
             num_workers=self.num_workers,
         )
 
@@ -523,7 +523,7 @@ class ChesapeakeCVPRDataModule(LightningDataModule):
         return DataLoader(
             self.val_dataset,
             batch_size=self.batch_size,
-            sampler=sampler,  # type: ignore[arg-type]
+            sampler=sampler,
             num_workers=self.num_workers,
         )
 
@@ -537,6 +537,6 @@ class ChesapeakeCVPRDataModule(LightningDataModule):
         return DataLoader(
             self.test_dataset,
             batch_size=self.batch_size,
-            sampler=sampler,  # type: ignore[arg-type]
+            sampler=sampler,
             num_workers=self.num_workers,
         )

torchgeo/trainers/naipchesapeake.py

@@ -325,7 +325,7 @@ class NAIPChesapeakeDataModule(pl.LightningDataModule):
         """Return a DataLoader for training."""
         return DataLoader(
             self.dataset,
-            batch_sampler=self.train_sampler,  # type: ignore[arg-type]
+            batch_sampler=self.train_sampler,
             num_workers=self.num_workers,
         )
 
@@ -334,7 +334,7 @@ class NAIPChesapeakeDataModule(pl.LightningDataModule):
         return DataLoader(
             self.dataset,
             batch_size=self.batch_size,
-            sampler=self.val_sampler,  # type: ignore[arg-type]
+            sampler=self.val_sampler,
             num_workers=self.num_workers,
         )
 
@@ -343,6 +343,6 @@ class NAIPChesapeakeDataModule(pl.LightningDataModule):
         return DataLoader(
             self.test_dataset,
             batch_size=self.batch_size,
-            sampler=self.test_sampler,  # type: ignore[arg-type]
+            sampler=self.test_sampler,
             num_workers=self.num_workers,
         )