Add latest mobilenetv4 and baseline updates for mobilenetv1 and efficientnet_b0 weights

Author: rwightman

timm/models/efficientnet.py

@@ -1238,9 +1238,17 @@ def _cfg(url='', **kwargs):
         url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mnasnet_small_lamb-aff75073.pth',
         hf_hub_id='timm/'),
 
-    'mobilenet_100.untrained': _cfg(),
-    'mobilenet_100h.untrained': _cfg(),
-    'mobilenet_125.untrained': _cfg(),
+    'mobilenetv1_100.ra4_e3600_r224_in1k': _cfg(
+        hf_hub_id='timm/',
+        mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
+        test_input_size=(3, 256, 256), test_crop_pct=0.95,
+    ),
+    'mobilenetv1_100h.ra4_e3600_r224_in1k': _cfg(
+        hf_hub_id='timm/',
+        mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
+        test_input_size=(3, 256, 256), test_crop_pct=0.95,
+    ),
+    'mobilenetv1_125.untrained': _cfg(),
 
     'mobilenetv2_035.untrained': _cfg(),
     'mobilenetv2_050.lamb_in1k': _cfg(
@@ -1275,22 +1283,27 @@ def _cfg(url='', **kwargs):
     'efficientnet_b0.ra_in1k': _cfg(
         url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_b0_ra-3dd342df.pth',
         hf_hub_id='timm/'),
+    'efficientnet_b0.ra4_e3600_r224_in1k': _cfg(
+        hf_hub_id='timm/',
+        mean=IMAGENET_INCEPTION_MEAN, std=IMAGENET_INCEPTION_STD,
+        crop_pct=0.9, test_input_size=(3, 256, 256), test_crop_pct=1.0
+    ),
     'efficientnet_b1.ft_in1k': _cfg(
         url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_b1-533bc792.pth',
         hf_hub_id='timm/',
-        test_input_size=(3, 256, 256), crop_pct=1.0),
+        test_input_size=(3, 256, 256), test_crop_pct=1.0),
     'efficientnet_b2.ra_in1k': _cfg(
         url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_b2_ra-bcdf34b7.pth',
         hf_hub_id='timm/',
-        input_size=(3, 256, 256), pool_size=(8, 8), test_input_size=(3, 288, 288), crop_pct=1.0),
+        input_size=(3, 256, 256), pool_size=(8, 8), test_input_size=(3, 288, 288), test_crop_pct=1.0),
     'efficientnet_b3.ra2_in1k': _cfg(
         url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_b3_ra2-cf984f9c.pth',
         hf_hub_id='timm/',
-        input_size=(3, 288, 288), pool_size=(9, 9), test_input_size=(3, 320, 320), crop_pct=1.0),
+        input_size=(3, 288, 288), pool_size=(9, 9), test_input_size=(3, 320, 320), test_crop_pct=1.0),
     'efficientnet_b4.ra2_in1k': _cfg(
         url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_b4_ra2_320-7eb33cd5.pth',
         hf_hub_id='timm/',
-        input_size=(3, 320, 320), pool_size=(10, 10), test_input_size=(3, 384, 384), crop_pct=1.0),
+        input_size=(3, 320, 320), pool_size=(10, 10), test_input_size=(3, 384, 384), test_crop_pct=1.0),
     'efficientnet_b5.sw_in12k_ft_in1k': _cfg(
         hf_hub_id='timm/',
         input_size=(3, 448, 448), pool_size=(14, 14), crop_pct=1.0, crop_mode='squash'),
@@ -1826,23 +1839,23 @@ def mnasnet_small(pretrained=False, **kwargs) -> EfficientNet:
 
 
 @register_model
-def mobilenet_100(pretrained=False, **kwargs) -> EfficientNet:
+def mobilenetv1_100(pretrained=False, **kwargs) -> EfficientNet:
     """ MobileNet V1 """
-    model = _gen_mobilenet_v1('mobilenet_100', 1.0, pretrained=pretrained, **kwargs)
+    model = _gen_mobilenet_v1('mobilenetv1_100', 1.0, pretrained=pretrained, **kwargs)
     return model
 
 
 @register_model
-def mobilenet_100h(pretrained=False, **kwargs) -> EfficientNet:
+def mobilenetv1_100h(pretrained=False, **kwargs) -> EfficientNet:
     """ MobileNet V1 """
-    model = _gen_mobilenet_v1('mobilenet_100h', 1.0, head_conv=True, pretrained=pretrained, **kwargs)
+    model = _gen_mobilenet_v1('mobilenetv1_100h', 1.0, head_conv=True, pretrained=pretrained, **kwargs)
     return model
 
 
 @register_model
-def mobilenet_125(pretrained=False, **kwargs) -> EfficientNet:
+def mobilenetv1_125(pretrained=False, **kwargs) -> EfficientNet:
     """ MobileNet V1 """
-    model = _gen_mobilenet_v1('mobilenet_125', 1.25, pretrained=pretrained, **kwargs)
+    model = _gen_mobilenet_v1('mobilenetv1_125', 1.25, pretrained=pretrained, **kwargs)
     return model
 
 

timm/models/mobilenetv3.py

@@ -1018,6 +1018,10 @@ def _cfg(url: str = '', **kwargs):
         input_size=(3, 256, 256), pool_size=(8, 8),
         crop_pct=0.95, test_input_size=(3, 320, 320), test_crop_pct=1.0, interpolation='bicubic'),
 
+    'mobilenetv4_hybrid_medium.e200_r256_in12k_ft_in1k': _cfg(
+        hf_hub_id='timm/',
+        input_size=(3, 256, 256), pool_size=(12, 12),
+        crop_pct=0.95, test_input_size=(3, 320, 320), test_crop_pct=1.0, interpolation='bicubic'),
     'mobilenetv4_hybrid_medium.ix_e550_r256_in1k': _cfg(
         hf_hub_id='timm/',
         input_size=(3, 256, 256), pool_size=(8, 8),
@@ -1029,6 +1033,11 @@ def _cfg(url: str = '', **kwargs):
     'mobilenetv4_hybrid_medium.e500_r224_in1k': _cfg(
         hf_hub_id='timm/',
         crop_pct=0.95, test_input_size=(3, 256, 256), test_crop_pct=1.0, interpolation='bicubic'),
+    'mobilenetv4_hybrid_medium.e200_r256_in12k': _cfg(
+        hf_hub_id='timm/',
+        num_classes=11821,
+        input_size=(3, 256, 256), pool_size=(12, 12),
+        crop_pct=0.95, test_input_size=(3, 320, 320), test_crop_pct=1.0, interpolation='bicubic'),
     'mobilenetv4_hybrid_large.ix_e600_r384_in1k': _cfg(
         hf_hub_id='timm/',
         input_size=(3, 384, 384), pool_size=(12, 12),
@@ -1045,12 +1054,21 @@ def _cfg(url: str = '', **kwargs):
     'mobilenetv4_conv_blur_medium.e500_r224_in1k': _cfg(
         hf_hub_id='timm/',
         crop_pct=0.95, test_input_size=(3, 256, 256), test_crop_pct=1.0, interpolation='bicubic'),
+    'mobilenetv4_conv_aa_large.e230_r448_in12k_ft_in1k': _cfg(
+        hf_hub_id='timm/',
+        input_size=(3, 448, 448), pool_size=(14, 14),
+        crop_pct=0.95, test_input_size=(3, 544, 544), test_crop_pct=1.0, interpolation='bicubic'),
+    'mobilenetv4_conv_aa_large.e230_r384_in12k_ft_in1k': _cfg(
+        hf_hub_id='timm/',
+        input_size=(3, 384, 384), pool_size=(12, 12),
+        crop_pct=0.95, test_input_size=(3, 480, 480), test_crop_pct=1.0, interpolation='bicubic'),
     'mobilenetv4_conv_aa_large.e600_r384_in1k': _cfg(
-        # hf_hub_id='timm/',
+        hf_hub_id='timm/',
         input_size=(3, 384, 384), pool_size=(12, 12),
-        crop_pct=0.95, test_input_size=(3, 448, 448), test_crop_pct=1.0, interpolation='bicubic'),
-    'mobilenetv4_conv_blur_large.e600_r384_in1k': _cfg(
-        # hf_hub_id='timm/',
+        crop_pct=0.95, test_input_size=(3, 480, 480), test_crop_pct=1.0, interpolation='bicubic'),
+    'mobilenetv4_conv_aa_large.e230_r384_in12k': _cfg(
+        hf_hub_id='timm/',
+        num_classes=11821,
         input_size=(3, 384, 384), pool_size=(12, 12),
         crop_pct=0.95, test_input_size=(3, 448, 448), test_crop_pct=1.0, interpolation='bicubic'),
     'mobilenetv4_hybrid_medium_075.untrained': _cfg(
@@ -1271,13 +1289,6 @@ def mobilenetv4_conv_aa_large(pretrained: bool = False, **kwargs) -> MobileNetV3
     return model
 
 
-@register_model
-def mobilenetv4_conv_blur_large(pretrained: bool = False, **kwargs) -> MobileNetV3:
-    """ MobileNet V4 Conv w/ Blur AA """
-    model = _gen_mobilenet_v4('mobilenetv4_conv_blur_large', 1.0, pretrained=pretrained, aa_layer='blurpc', **kwargs)
-    return model
-
-
 @register_model
 def mobilenetv4_hybrid_medium_075(pretrained: bool = False, **kwargs) -> MobileNetV3:
     """ MobileNet V4 Hybrid """