Improve filtering behaviour for tag + non-tagged model wildcard consistency.

rwightman · rwightman · commit b3e816d6d71e · 2023-03-22T10:21:22.000-07:00
diff --git a/tests/test_models.py b/tests/test_models.py
@@ -52,7 +52,7 @@
         '*efficientnet_l2*', '*resnext101_32x48d', '*in21k', '*152x4_bitm', '*101x3_bitm', '*50x3_bitm',
         '*nfnet_f3*', '*nfnet_f4*', '*nfnet_f5*', '*nfnet_f6*', '*nfnet_f7*', '*efficientnetv2_xl*',
         '*resnetrs350*', '*resnetrs420*', 'xcit_large_24_p8*', 'vit_huge*', 'vit_gi*', 'swin*huge*',
-        'swin*giant*', 'convnextv2_huge*', 'maxvit_xlarge*', 'davit_giant', 'davit_huge']
+        'swin*giant*', 'convnextv2_huge*', 'maxvit_xlarge*', 'davit_giant', 'davit_huge', 'regnet*1280', 'regnet*2560']
     NON_STD_EXCLUDE_FILTERS = ['vit_huge*', 'vit_gi*',  'eva_giant*']
 else:
     EXCLUDE_FILTERS = []
diff --git a/timm/models/_registry.py b/timm/models/_registry.py
@@ -167,6 +167,15 @@ def _natural_key(string_: str) -> List[Union[int, str]]:
     return [int(s) if s.isdigit() else s for s in re.split(r'(\d+)', string_.lower())]
 
 
+def _expand_filter(filter: str):
+    """ expand a 'base_filter' to 'base_filter.*' if no tag portion"""
+    filter_base, filter_tag = split_model_name_tag(filter)
+    if not filter_tag:
+        return ['.'.join([filter_base, '*']), filter]
+    else:
+        return [filter]
+
+
 def list_models(
         filter: Union[str, List[str]] = '',
         module: str = '',
@@ -193,6 +202,11 @@ def list_models(
         model_list('gluon_resnet*') -- returns all models starting with 'gluon_resnet'
         model_list('*resnext*, 'resnet') -- returns all models with 'resnext' in 'resnet' module
     """
+    if filter:
+        include_filters = filter if isinstance(filter, (tuple, list)) else [filter]
+    else:
+        include_filters = []
+
     if include_tags is None:
         # FIXME should this be default behaviour? or default to include_tags=True?
         include_tags = pretrained
@@ -206,10 +220,12 @@ def list_models(
         for m in all_models:
             models_with_tags.update(_model_with_tags[m])
         all_models = models_with_tags
+        # expand include and exclude filters to include a '.*' for proper match if no tags in filter
+        include_filters = [ef for f in include_filters for ef in _expand_filter(f)]
+        exclude_filters = [ef for f in exclude_filters for ef in _expand_filter(f)]
 
-    if filter:
+    if include_filters:
         models: Set[str] = set()
-        include_filters = filter if isinstance(filter, (tuple, list)) else [filter]
         for f in include_filters:
             include_models = fnmatch.filter(all_models, f)  # include these models
             if len(include_models):
diff --git a/validate.py b/validate.py
@@ -385,6 +385,9 @@ def _try_run(args, initial_batch_size):
     return results
 
 
+_NON_IN1K_FILTERS = ['*_in21k', '*_in22k', '*in12k', '*_dino', '*fcmae', '*seer']
+
+
 def main():
     setup_default_logging()
     args = parser.parse_args()
@@ -400,11 +403,17 @@ def main():
         if args.model == 'all':
             # validate all models in a list of names with pretrained checkpoints
             args.pretrained = True
-            model_names = list_models('convnext*', pretrained=True, exclude_filters=['*_in21k', '*_in22k', '*in12k', '*_dino', '*fcmae'])
+            model_names = list_models(
+                pretrained=True,
+                exclude_filters=_NON_IN1K_FILTERS,
+            )
             model_cfgs = [(n, '') for n in model_names]
         elif not is_model(args.model):
             # model name doesn't exist, try as wildcard filter
-            model_names = list_models(args.model, pretrained=True)
+            model_names = list_models(
+                args.model,
+                pretrained=True,
+            )
             model_cfgs = [(n, '') for n in model_names]
 
         if not model_cfgs and os.path.isfile(args.model):
