torch.hub: add support for DOFA and Swin models (#2052)

* torch.hub: add support for DOFA and Swin models

* Fix tests

* Add *args support to DOFA to match other models

Author: adamjstewart

hubconf.py

@@ -7,8 +7,22 @@
 * https://pytorch.org/docs/stable/hub.html
 """
 
-from torchgeo.models import resnet18, resnet50, vit_small_patch16_224
+from torchgeo.models import (
+    dofa_base_patch16_224,
+    dofa_large_patch16_224,
+    resnet18,
+    resnet50,
+    swin_v2_b,
+    vit_small_patch16_224,
+)
 
-__all__ = ('resnet18', 'resnet50', 'vit_small_patch16_224')
+__all__ = (
+    'dofa_base_patch16_224',
+    'dofa_large_patch16_224',
+    'resnet18',
+    'resnet50',
+    'swin_v2_b',
+    'vit_small_patch16_224',
+)
 
 dependencies = ['timm']

tests/models/test_api.py

@@ -9,10 +9,14 @@
 from torchvision.models._api import WeightsEnum
 
 from torchgeo.models import (
+    DOFABase16_Weights,
+    DOFALarge16_Weights,
     ResNet18_Weights,
     ResNet50_Weights,
     Swin_V2_B_Weights,
     ViTSmall16_Weights,
+    dofa_base_patch16_224,
+    dofa_large_patch16_224,
     get_model,
     get_model_weights,
     get_weight,
@@ -23,8 +27,22 @@
     vit_small_patch16_224,
 )
 
-builders = [resnet18, resnet50, vit_small_patch16_224, swin_v2_b]
-enums = [ResNet18_Weights, ResNet50_Weights, ViTSmall16_Weights, Swin_V2_B_Weights]
+builders = [
+    dofa_base_patch16_224,
+    dofa_large_patch16_224,
+    resnet18,
+    resnet50,
+    swin_v2_b,
+    vit_small_patch16_224,
+]
+enums = [
+    DOFABase16_Weights,
+    DOFALarge16_Weights,
+    ResNet18_Weights,
+    ResNet50_Weights,
+    Swin_V2_B_Weights,
+    ViTSmall16_Weights,
+]
 
 
 @pytest.mark.parametrize('builder', builders)

torchgeo/models/api.py

@@ -16,26 +16,38 @@
 import torch.nn as nn
 from torchvision.models._api import WeightsEnum
 
+from .dofa import (
+    DOFABase16_Weights,
+    DOFALarge16_Weights,
+    dofa_base_patch16_224,
+    dofa_large_patch16_224,
+)
 from .resnet import ResNet18_Weights, ResNet50_Weights, resnet18, resnet50
 from .swin import Swin_V2_B_Weights, swin_v2_b
 from .vit import ViTSmall16_Weights, vit_small_patch16_224
 
 _model = {
+    'dofa_base_patch16_224': dofa_base_patch16_224,
+    'dofa_large_patch16_224': dofa_large_patch16_224,
     'resnet18': resnet18,
     'resnet50': resnet50,
-    'vit_small_patch16_224': vit_small_patch16_224,
     'swin_v2_b': swin_v2_b,
+    'vit_small_patch16_224': vit_small_patch16_224,
 }
 
 _model_weights = {
+    dofa_base_patch16_224: DOFABase16_Weights,
+    dofa_large_patch16_224: DOFALarge16_Weights,
     resnet18: ResNet18_Weights,
     resnet50: ResNet50_Weights,
-    vit_small_patch16_224: ViTSmall16_Weights,
     swin_v2_b: Swin_V2_B_Weights,
+    vit_small_patch16_224: ViTSmall16_Weights,
+    'dofa_base_patch16_224': DOFABase16_Weights,
+    'dofa_large_patch16_224': DOFALarge16_Weights,
     'resnet18': ResNet18_Weights,
     'resnet50': ResNet50_Weights,
-    'vit_small_patch16_224': ViTSmall16_Weights,
     'swin_v2_b': Swin_V2_B_Weights,
+    'vit_small_patch16_224': ViTSmall16_Weights,
 }
 
 

torchgeo/models/dofa.py

@@ -415,7 +415,7 @@ class DOFALarge16_Weights(WeightsEnum):  # type: ignore[misc]
     )
 
 
-def dofa_small_patch16_224(**kwargs: Any) -> DOFA:
+def dofa_small_patch16_224(*args: Any, **kwargs: Any) -> DOFA:
     """Dynamic One-For-All (DOFA) small patch size 16 model.
 
     If you use this model in your research, please cite the following paper:
@@ -425,17 +425,19 @@ def dofa_small_patch16_224(**kwargs: Any) -> DOFA:
     .. versionadded:: 0.6
 
     Args:
+        *args: Additional arguments to pass to :class:`DOFA`.
         **kwargs: Additional keywork arguments to pass to :class:`DOFA`.
 
     Returns:
         A DOFA small 16 model.
     """
-    model = DOFA(patch_size=16, embed_dim=384, depth=12, num_heads=6, **kwargs)
+    kwargs |= {'patch_size': 16, 'embed_dim': 384, 'depth': 12, 'num_heads': 6}
+    model = DOFA(*args, **kwargs)
     return model
 
 
 def dofa_base_patch16_224(
-    weights: DOFABase16_Weights | None = None, **kwargs: Any
+    weights: DOFABase16_Weights | None = None, *args: Any, **kwargs: Any
 ) -> DOFA:
     """Dynamic One-For-All (DOFA) base patch size 16 model.
 
@@ -447,12 +449,14 @@ def dofa_base_patch16_224(
 
     Args:
         weights: Pre-trained model weights to use.
+        *args: Additional arguments to pass to :class:`DOFA`.
         **kwargs: Additional keywork arguments to pass to :class:`DOFA`.
 
     Returns:
         A DOFA base 16 model.
     """
-    model = DOFA(patch_size=16, embed_dim=768, depth=12, num_heads=12, **kwargs)
+    kwargs |= {'patch_size': 16, 'embed_dim': 768, 'depth': 12, 'num_heads': 12}
+    model = DOFA(*args, **kwargs)
 
     if weights:
         missing_keys, unexpected_keys = model.load_state_dict(
@@ -471,7 +475,7 @@ def dofa_base_patch16_224(
 
 
 def dofa_large_patch16_224(
-    weights: DOFALarge16_Weights | None = None, **kwargs: Any
+    weights: DOFALarge16_Weights | None = None, *args: Any, **kwargs: Any
 ) -> DOFA:
     """Dynamic One-For-All (DOFA) large patch size 16 model.
 
@@ -483,12 +487,14 @@ def dofa_large_patch16_224(
 
     Args:
         weights: Pre-trained model weights to use.
+        *args: Additional arguments to pass to :class:`DOFA`.
         **kwargs: Additional keywork arguments to pass to :class:`DOFA`.
 
     Returns:
         A DOFA large 16 model.
     """
-    model = DOFA(patch_size=16, embed_dim=1024, depth=24, num_heads=16, **kwargs)
+    kwargs |= {'patch_size': 16, 'embed_dim': 1024, 'depth': 24, 'num_heads': 16}
+    model = DOFA(*args, **kwargs)
 
     if weights:
         missing_keys, unexpected_keys = model.load_state_dict(
@@ -506,7 +512,7 @@ def dofa_large_patch16_224(
     return model
 
 
-def dofa_huge_patch16_224(**kwargs: Any) -> DOFA:
+def dofa_huge_patch16_224(*args: Any, **kwargs: Any) -> DOFA:
     """Dynamic One-For-All (DOFA) huge patch size 16 model.
 
     If you use this model in your research, please cite the following paper:
@@ -516,10 +522,12 @@ def dofa_huge_patch16_224(**kwargs: Any) -> DOFA:
     .. versionadded:: 0.6
 
     Args:
+        *args: Additional arguments to pass to :class:`DOFA`.
         **kwargs: Additional keywork arguments to pass to :class:`DOFA`.
 
     Returns:
         A DOFA huge 16 model.
     """
-    model = DOFA(patch_size=14, embed_dim=1280, depth=32, num_heads=16, **kwargs)
+    kwargs |= {'patch_size': 14, 'embed_dim': 1280, 'depth': 32, 'num_heads': 16}
+    model = DOFA(*args, **kwargs)
     return model