up

timm/models/helpers.py

@@ -200,12 +200,17 @@ def load_pretrained(model, cfg=None, num_classes=1000, in_chans=3, filter_fn=Non
                 _logger.warning(
                     f'Unable to convert pretrained {input_conv_name} weights, using random init for this layer.')
 
-    classifier_name = cfg['classifier']
+    classifiers = cfg['classifier']
     label_offset = cfg.get('label_offset', 0)
     if num_classes != cfg['num_classes']:
         # completely discard fully connected if model num_classes doesn't match pretrained weights
-        del state_dict[classifier_name + '.weight']
+        if isinstance(classifiers, str):
-        del state_dict[classifier_name + '.bias']
+            classifiers = (classifiers,)
+        for classifier_name in classifiers:
+            classifier_weight = classifier_name + '.weight'
+            classifier_bias = classifier_name + '.bias'
+            del state_dict[classifier_weight]
+            del state_dict[classifier_bias]
         strict = False
     elif label_offset > 0:
         # special case for pretrained weights with an extra background class in pretrained weights