Consistent license handling throughout timm

timm/models/_hub.py

@@ -32,9 +32,9 @@
 try:
     from huggingface_hub import (
         create_repo, get_hf_file_metadata,
-        hf_hub_download, hf_hub_url,
+        hf_hub_download, hf_hub_url, model_info,
         repo_type_and_id_from_hf_id, upload_folder)
-    from huggingface_hub.utils import EntryNotFoundError
+    from huggingface_hub.utils import EntryNotFoundError, RepositoryNotFoundError
     hf_hub_download = partial(hf_hub_download, library_name="timm", library_version=__version__)
     _has_hf_hub = True
 except ImportError:
@@ -540,3 +540,56 @@ def _get_safe_alternatives(filename: str) -> Iterable[str]:
         yield HF_OPEN_CLIP_SAFE_WEIGHTS_NAME
     if filename not in (HF_WEIGHTS_NAME, HF_OPEN_CLIP_WEIGHTS_NAME) and filename.endswith(".bin"):
         yield filename[:-4] + ".safetensors"
+
+
+def _get_license_from_hf_hub(model_id: str | None, hf_hub_id: str | None) -> str | None:
+    """Retrieve license information for a model from Hugging Face Hub.
+
+    Fetches the license field from the model card metadata on Hugging Face Hub
+    for the specified model. Returns None if the model is not found, if
+    huggingface_hub is not installed, or if the model is marked as "untrained".
+
+    Args:
+        model_id: The model identifier/name. In the case of None we assume an untrained model.
+        hf_hub_id: The Hugging Face Hub organization/user ID. If it is None,
+            we will return None as we cannot infer the license terms.
+
+    Returns:
+        The license string in lowercase if found, None otherwise.
+
+    Note:
+        Requires huggingface_hub package to be installed. Will log a warning
+        and return None if the package is not available.
+    """
+    if not has_hf_hub(True):
+        msg = "For updated license information run `pip install huggingface_hub`."
+        _logger.warning(msg=msg)
+        return None
+
+    if not (model_id and hf_hub_id):
+        return None
+
+    repo_id: str = hf_hub_id + model_id
+
+    try:
+        info = model_info(repo_id=repo_id)
+
+    except RepositoryNotFoundError:
+        msg = f"Repository {repo_id} was not found. Manual inspection of license needed."
+        _logger.warning(msg=msg)
+        return None
+
+    except Exception as _:
+        msg = f"Error for {repo_id}. Manual inspection of license needed."
+        _logger.warning(msg=msg)
+        return None
+
+    license = info.card_data.get("license").lower() if info.card_data else None
+
+    if license == 'other':
+        name = info.card_data.get("license_name", None)
+
+        if name is not None:
+            return name
+
+    return license

timm/models/beit.py

@@ -738,6 +738,7 @@ def _cfg(url: str = '', **kwargs) -> Dict[str, Any]:
         'crop_pct': .9, 'interpolation': 'bicubic', 'fixed_input_size': True,
         'mean': (0.5, 0.5, 0.5), 'std': (0.5, 0.5, 0.5),
         'first_conv': 'patch_embed.proj', 'classifier': 'head',
+        'license': 'apache-2.0',
         **kwargs
     }
 

timm/models/byoanet.py

@@ -297,7 +297,7 @@ def _cfg(url: str = '', **kwargs) -> Dict[str, Any]:
         'crop_pct': 0.95, 'interpolation': 'bicubic',
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
         'first_conv': 'stem.conv1.conv', 'classifier': 'head.fc',
-        'fixed_input_size': False, 'min_input_size': (3, 224, 224),
+        'fixed_input_size': False, 'min_input_size': (3, 224, 224), 'license': 'apache-2.0',
         **kwargs
     }
 

timm/models/byobnet.py

@@ -2279,6 +2279,7 @@ def _cfg(url: str = '', **kwargs) -> Dict[str, Any]:
         'crop_pct': 0.875, 'interpolation': 'bilinear',
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
         'first_conv': 'stem.conv', 'classifier': 'head.fc',
+        'license': 'apache-2.0',
         **kwargs
     }
 
@@ -2298,6 +2299,7 @@ def _cfgr(url: str = '', **kwargs) -> Dict[str, Any]:
         'crop_pct': 0.9, 'interpolation': 'bicubic',
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
         'first_conv': 'stem.conv1.conv', 'classifier': 'head.fc',
+        'license': 'apache-2.0',
         **kwargs
     }
 
@@ -2451,26 +2453,31 @@ def _cfgr(url: str = '', **kwargs) -> Dict[str, Any]:
         hf_hub_id='timm/',
         crop_pct=0.875,
         first_conv=('stem.conv_kxk.0.conv', 'stem.conv_scale.conv'),
+        license='mobileone-license',
     ),
     'mobileone_s1.apple_in1k': _cfg(
         hf_hub_id='timm/',
         crop_pct=0.9,
         first_conv=('stem.conv_kxk.0.conv', 'stem.conv_scale.conv'),
+        license='mobileone-license',
     ),
     'mobileone_s2.apple_in1k': _cfg(
         hf_hub_id='timm/',
         crop_pct=0.9,
         first_conv=('stem.conv_kxk.0.conv', 'stem.conv_scale.conv'),
+        license='mobileone-license',
     ),
     'mobileone_s3.apple_in1k': _cfg(
         hf_hub_id='timm/',
         crop_pct=0.9,
         first_conv=('stem.conv_kxk.0.conv', 'stem.conv_scale.conv'),
+        license='mobileone-license',
     ),
     'mobileone_s4.apple_in1k': _cfg(
         hf_hub_id='timm/',
         crop_pct=0.9,
         first_conv=('stem.conv_kxk.0.conv', 'stem.conv_scale.conv'),
+        license='mobileone-license',
     ),
 
     # original attention pool head variants
@@ -2479,90 +2486,106 @@ def _cfgr(url: str = '', **kwargs) -> Dict[str, Any]:
         num_classes=1024, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         fixed_input_size=True, input_size=(3, 224, 224), pool_size=(7, 7),
         classifier='head.proj',
+        license='mit',
     ),
     'resnet101_clip.openai': _cfgr(
         hf_hub_id='timm/',
         num_classes=512, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         fixed_input_size=True, input_size=(3, 224, 224), pool_size=(7, 7),
         classifier='head.proj',
+        license='mit',
     ),
     'resnet50x4_clip.openai': _cfgr(
         hf_hub_id='timm/',
         num_classes=640, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         fixed_input_size=True, input_size=(3, 288, 288), pool_size=(9, 9),
         classifier='head.proj',
+        license='mit',
     ),
     'resnet50x16_clip.openai': _cfgr(
         hf_hub_id='timm/',
         num_classes=768, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         fixed_input_size=True, input_size=(3, 384, 384), pool_size=(12, 12),
         classifier='head.proj',
+        license='mit',
     ),
     'resnet50x64_clip.openai': _cfgr(
         hf_hub_id='timm/',
         num_classes=1024, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         fixed_input_size=True, input_size=(3, 448, 448), pool_size=(14, 14),
         classifier='head.proj',
+        license='mit',
     ),
     'resnet50_clip.cc12m': _cfgr(
         hf_hub_id='timm/',
         num_classes=1024, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         fixed_input_size=True, input_size=(3, 224, 224), pool_size=(7, 7),
         classifier='head.proj',
+        license='mit',
     ),
     'resnet50_clip.yfcc15m': _cfgr(
         hf_hub_id='timm/',
         num_classes=1024, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         fixed_input_size=True, input_size=(3, 224, 224), pool_size=(7, 7),
         classifier='head.proj',
+        license='mit',
     ),
     'resnet101_clip.yfcc15m': _cfgr(
         hf_hub_id='timm/',
         num_classes=512, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         fixed_input_size=True, input_size=(3, 224, 224), pool_size=(7, 7),
         classifier='head.proj',
+        license='mit',
     ),
 
     # avg-pool w/ optional standard classifier head variants
     'resnet50_clip_gap.openai': _cfgr(
         hf_hub_id='timm/',
         num_classes=0, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         input_size=(3, 224, 224), pool_size=(7, 7),
+        license='mit',
     ),
     'resnet101_clip_gap.openai': _cfgr(
         hf_hub_id='timm/',
         num_classes=0, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         input_size=(3, 224, 224), pool_size=(7, 7),
+        license='mit',
     ),
     'resnet50x4_clip_gap.openai': _cfgr(
         hf_hub_id='timm/',
         num_classes=0, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         input_size=(3, 288, 288), pool_size=(9, 9),
+        license='mit',
     ),
     'resnet50x16_clip_gap.openai': _cfgr(
         hf_hub_id='timm/',
         num_classes=0, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         input_size=(3, 384, 384), pool_size=(12, 12),
+        license='mit',
     ),
     'resnet50x64_clip_gap.openai': _cfgr(
         hf_hub_id='timm/',
         num_classes=0, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         input_size=(3, 448, 448), pool_size=(14, 14),
+        license='mit',
     ),
     'resnet50_clip_gap.cc12m': _cfgr(
         hf_hub_id='timm/',
         num_classes=0, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         input_size=(3, 224, 224), pool_size=(7, 7),
+        license='mit',
     ),
     'resnet50_clip_gap.yfcc15m': _cfgr(
         hf_hub_id='timm/',
         num_classes=0, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         input_size=(3, 224, 224), pool_size=(7, 7),
+        license='mit',
     ),
     'resnet101_clip_gap.yfcc15m': _cfgr(
         hf_hub_id='timm/',
         num_classes=0, mean=OPENAI_CLIP_MEAN, std=OPENAI_CLIP_STD,
         input_size=(3, 224, 224), pool_size=(7, 7),
+        license='mit',
     ),
 
     'resnet50_mlp.untrained': _cfgr(

timm/models/cait.py

@@ -472,6 +472,7 @@ def _cfg(url='', **kwargs):
         'crop_pct': 1.0, 'interpolation': 'bicubic', 'fixed_input_size': True,
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
         'first_conv': 'patch_embed.proj', 'classifier': 'head',
+        'license': 'apache-2.0',
         **kwargs
     }
 

timm/models/coat.py

@@ -720,6 +720,7 @@ def _cfg_coat(url='', **kwargs):
         'crop_pct': .9, 'interpolation': 'bicubic', 'fixed_input_size': True,
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
         'first_conv': 'patch_embed1.proj', 'classifier': 'head',
+        'license': 'apache-2.0',
         **kwargs
     }
 
@@ -800,4 +801,4 @@ def coat_lite_medium_384(pretrained=False, **kwargs) -> CoaT:
     model_cfg = dict(
         img_size=384, patch_size=4, embed_dims=[128, 256, 320, 512], serial_depths=[3, 6, 10, 8])
     model = _create_coat('coat_lite_medium_384', pretrained=pretrained, **dict(model_cfg, **kwargs))
-    return model
+    return model

timm/models/convit.py

@@ -391,7 +391,7 @@ def _cfg(url='', **kwargs):
         'url': url,
         'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None,
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD, 'fixed_input_size': True,
-        'first_conv': 'patch_embed.proj', 'classifier': 'head',
+        'first_conv': 'patch_embed.proj', 'classifier': 'head', 'license': 'apache-2.0',
         **kwargs
     }
 

timm/models/convmixer.py

@@ -115,7 +115,7 @@ def _cfg(url='', **kwargs):
         'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None,
         'crop_pct': .96, 'interpolation': 'bicubic',
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD, 'classifier': 'head',
-        'first_conv': 'stem.0',
+        'first_conv': 'stem.0', 'license': 'mit',
         **kwargs
     }
 
@@ -143,4 +143,4 @@ def convmixer_768_32(pretrained=False, **kwargs) -> ConvMixer:
 @register_model
 def convmixer_1024_20_ks9_p14(pretrained=False, **kwargs) -> ConvMixer:
     model_args = dict(dim=1024, depth=20, kernel_size=9, patch_size=14, **kwargs)
-    return _create_convmixer('convmixer_1024_20_ks9_p14', pretrained, **model_args)
+    return _create_convmixer('convmixer_1024_20_ks9_p14', pretrained, **model_args)

timm/models/convnext.py

@@ -665,7 +665,7 @@ def _cfg(url='', **kwargs):
         'crop_pct': 0.875, 'interpolation': 'bicubic',
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
         'first_conv': 'stem.0', 'classifier': 'head.fc',
-        **kwargs
+        'license': 'apache-2.0', **kwargs
     }
 
 
@@ -1067,25 +1067,25 @@ def _cfgv2(url='', **kwargs):
         hf_hub_id='timm/',
         crop_pct=1.0,
         num_classes=0,
-        license='dinov3',
+        license='dinov3-license',
     ),
     'convnext_small.dinov3_lvd1689m': _cfg(
         hf_hub_id='timm/',
         crop_pct=1.0,
         num_classes=0,
-        license='dinov3',
+        license='dinov3-license',
     ),
     'convnext_base.dinov3_lvd1689m': _cfg(
         hf_hub_id='timm/',
         crop_pct=1.0,
         num_classes=0,
-        license='dinov3',
+        license='dinov3-license',
     ),
     'convnext_large.dinov3_lvd1689m': _cfg(
         hf_hub_id='timm/',
         crop_pct=1.0,
         num_classes=0,
-        license='dinov3',
+        license='dinov3-license',
     ),
 
     "test_convnext.r160_in1k": _cfg(

timm/models/crossvit.py

@@ -489,6 +489,7 @@ def _cfg(url='', **kwargs):
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD, 'fixed_input_size': True,
         'first_conv': ('patch_embed.0.proj', 'patch_embed.1.proj'),
         'classifier': ('head.0', 'head.1'),
+        'license': 'apache-2.0',
         **kwargs
     }
 

timm/models/cspnet.py

@@ -917,7 +917,7 @@ def _cfg(url='', **kwargs):
         'num_classes': 1000, 'input_size': (3, 256, 256), 'pool_size': (8, 8),
         'crop_pct': 0.887, 'interpolation': 'bilinear',
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
-        'first_conv': 'stem.conv1.conv', 'classifier': 'head.fc',
+        'first_conv': 'stem.conv1.conv', 'classifier': 'head.fc', 'license': 'apache-2.0',
         **kwargs
     }
 
@@ -1111,4 +1111,4 @@ def cs3edgenet_x(pretrained=False, **kwargs) -> CspNet:
 
 @register_model
 def cs3se_edgenet_x(pretrained=False, **kwargs) -> CspNet:
-    return _create_cspnet('cs3se_edgenet_x', pretrained=pretrained, **kwargs)
+    return _create_cspnet('cs3se_edgenet_x', pretrained=pretrained, **kwargs)

timm/models/davit.py

@@ -806,6 +806,7 @@ def _cfg(url='', **kwargs):
         'crop_pct': 0.95, 'interpolation': 'bicubic',
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
         'first_conv': 'stem.conv', 'classifier': 'head.fc',
+        'license': 'apache-2.0',
         **kwargs
     }
 

timm/models/deit.py

@@ -140,6 +140,7 @@ def _cfg(url='', **kwargs):
         'crop_pct': .9, 'interpolation': 'bicubic', 'fixed_input_size': True,
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
         'first_conv': 'patch_embed.proj', 'classifier': 'head',
+        'license': 'apache-2.0',
         **kwargs
     }
 

timm/models/densenet.py

@@ -472,7 +472,8 @@ def _cfg(url: str = '', **kwargs) -> Dict[str, Any]:
         'url': url, 'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': (7, 7),
         'crop_pct': 0.875, 'interpolation': 'bicubic',
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
-        'first_conv': 'features.conv0', 'classifier': 'classifier', **kwargs,
+        'first_conv': 'features.conv0', 'classifier': 'classifier', 'license': 'apache-2.0',
+        **kwargs,
     }
 
 

timm/models/dla.py

@@ -399,7 +399,7 @@ def _cfg(url='', **kwargs):
         'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': (7, 7),
         'crop_pct': 0.875, 'interpolation': 'bilinear',
         'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
-        'first_conv': 'base_layer.0', 'classifier': 'fc',
+        'first_conv': 'base_layer.0', 'classifier': 'fc', 'license': 'bsd-3-clause',
         **kwargs
     }
 
@@ -415,8 +415,8 @@ def _cfg(url='', **kwargs):
     'dla102x.in1k': _cfg(hf_hub_id='timm/'),
     'dla102x2.in1k': _cfg(hf_hub_id='timm/'),
     'dla169.in1k': _cfg(hf_hub_id='timm/'),
-    'dla60_res2net.in1k': _cfg(hf_hub_id='timm/'),
-    'dla60_res2next.in1k': _cfg(hf_hub_id='timm/'),
+    'dla60_res2net.in1k': _cfg(hf_hub_id='timm/', license='unknown'),
+    'dla60_res2next.in1k': _cfg(hf_hub_id='timm/', license='unknown'),
 })