Add features_only, other bits to mambaout, define different base alternatives

Author: rwightman

timm/models/mambaout.py

@@ -5,6 +5,7 @@
 MetaFormer (https://github.com/sail-sg/metaformer),
 InceptionNeXt (https://github.com/sail-sg/inceptionnext)
 """
+from collections import OrderedDict
 from typing import Optional
 
 import torch
@@ -120,7 +121,7 @@ class MlpHead(nn.Module):
 
     def __init__(
             self,
-            dim,
+            in_features,
             num_classes=1000,
             pool_type='avg',
             act_layer=nn.GELU,
@@ -130,27 +131,47 @@ def __init__(
             bias=True,
     ):
         super().__init__()
-        hidden_features = int(mlp_ratio * dim)
+        if mlp_ratio is not None:
+            hidden_size = int(mlp_ratio * in_features)
+        else:
+            hidden_size = None
         self.pool_type = pool_type
+        self.in_features = in_features
+        self.hidden_size = hidden_size or in_features
+
+        self.norm = norm_layer(in_features)
+        if hidden_size:
+            self.pre_logits = nn.Sequential(OrderedDict([
+                ('fc', nn.Linear(in_features, hidden_size)),
+                ('act', act_layer()),
+                ('norm', norm_layer(hidden_size))
+            ]))
+            self.num_features = hidden_size
+        else:
+            self.num_features = in_features
+            self.pre_logits = nn.Identity()
 
-        self.norm1 = norm_layer(dim)
-        self.fc1 = nn.Linear(dim, hidden_features, bias=bias)
-        self.act = act_layer()
-        self.norm2 = norm_layer(hidden_features)
-        self.fc2 = nn.Linear(hidden_features, num_classes, bias=bias)
+        self.fc = nn.Linear(hidden_size, num_classes, bias=bias)
         self.head_dropout = nn.Dropout(drop_rate)
 
+    def reset(self, num_classes: int, pool_type: Optional[str] = None, reset_other: bool = False):
+        if pool_type is not None:
+            self.pool_type = pool_type
+        if reset_other:
+            self.norm = nn.Identity()
+            self.pre_logits = nn.Identity()
+            self.num_features = self.in_features
+        self.fc = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity()
+
     def forward(self, x, pre_logits: bool = False):
         if self.pool_type == 'avg':
             x = x.mean((1, 2))
-        x = self.norm1(x)
-        x = self.fc1(x)
-        x = self.act(x)
-        x = self.norm2(x)
+        x = self.norm(x)
+        x = self.pre_logits(x)
         x = self.head_dropout(x)
         if pre_logits:
             return x
-        x = self.fc2(x)
+        x = self.fc(x)
         return x
 
 
@@ -284,6 +305,7 @@ def __init__(
             norm_layer=LayerNorm,
             act_layer=nn.GELU,
             conv_ratio=1.0,
+            expansion_ratio=8/3,
             kernel_size=7,
             stem_mid_norm=True,
             ls_init_value=None,
@@ -303,6 +325,7 @@ def __init__(
 
         num_stage = len(depths)
         self.num_stage = num_stage
+        self.feature_info = []
 
         self.stem = Stem(
             in_chans,
@@ -313,16 +336,20 @@ def __init__(
         )
         prev_dim = dims[0]
         dp_rates = [x.tolist() for x in torch.linspace(0, drop_path_rate, sum(depths)).split(depths)]
-        self.stages = nn.ModuleList()
         cur = 0
+        curr_stride = 4
+        self.stages = nn.Sequential()
         for i in range(num_stage):
             dim = dims[i]
+            stride = 2 if curr_stride == 2 or i > 0 else 1
+            curr_stride *= stride
             stage = MambaOutStage(
                 dim=prev_dim,
                 dim_out=dim,
                 depth=depths[i],
                 kernel_size=kernel_size,
                 conv_ratio=conv_ratio,
+                expansion_ratio=expansion_ratio,
                 downsample=downsample if i > 0 else '',
                 ls_init_value=ls_init_value,
                 norm_layer=norm_layer,
@@ -331,6 +358,8 @@ def __init__(
             )
             self.stages.append(stage)
             prev_dim = dim
+            # NOTE feature_info use currently assumes stage 0 == stride 1, rest are stride 2
+            self.feature_info += [dict(num_chs=prev_dim, reduction=curr_stride, module=f'stages.{i}')]
             cur += depths[i]
 
         if head_fn == 'default':
@@ -352,6 +381,8 @@ def __init__(
                 norm_layer=norm_layer,
                 drop_rate=drop_rate,
             )
+        self.num_features = prev_dim
+        self.hidden_size = self.head.num_features
 
         self.apply(self._init_weights)
 
@@ -362,13 +393,31 @@ def _init_weights(self, m):
                 nn.init.constant_(m.bias, 0)
 
     @torch.jit.ignore
-    def no_weight_decay(self):
-        return {}
+    def group_matcher(self, coarse=False):
+        return dict(
+            stem=r'^stem',
+            blocks=r'^stages\.(\d+)' if coarse else [
+                (r'^stages\.(\d+)\.downsample', (0,)),  # blocks
+                (r'^stages\.(\d+)\.blocks\.(\d+)', None),
+            ]
+        )
+
+    @torch.jit.ignore
+    def set_grad_checkpointing(self, enable=True):
+        for s in self.stages:
+            s.grad_checkpointing = enable
+
+    @torch.jit.ignore
+    def get_classifier(self) -> nn.Module:
+        return self.head.fc
+
+    def reset_classifier(self, num_classes: int, global_pool: Optional[str] = None):
+        self.num_classes = num_classes
+        self.head.reset(num_classes, global_pool)
 
     def forward_features(self, x):
         x = self.stem(x)
-        for s in self.stages:
-            x = s(x)
+        x = self.stages(x)
         return x
 
     def forward_head(self, x, pre_logits: bool = False):
@@ -391,10 +440,14 @@ def checkpoint_filter_fn(state_dict, model):
         k = k.replace('downsample_layers.0.', 'stem.')
         k = re.sub(r'stages.([0-9]+).([0-9]+)', r'stages.\1.blocks.\2', k)
         k = re.sub(r'downsample_layers.([0-9]+)', r'stages.\1.downsample', k)
+        # remap head names
         if k.startswith('norm.'):
-            k = k.replace('norm.', 'head.norm1.')
-        elif k.startswith('head.norm.'):
-            k = k.replace('head.norm.', 'head.norm2.')
+            # this is moving to head since it's after the pooling
+            k = k.replace('norm.', 'head.norm.')
+        elif k.startswith('head.'):
+            k = k.replace('head.fc1.', 'head.pre_logits.fc.')
+            k = k.replace('head.norm.', 'head.pre_logits.norm.')
+            k = k.replace('head.fc2.', 'head.fc.')
         out_dict[k] = v
 
     return out_dict
@@ -405,7 +458,7 @@ def _cfg(url='', **kwargs):
         'url': url,
         'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None,
         'crop_pct': 1.0, 'interpolation': 'bicubic',
-        'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD, 'classifier': 'head',
+        'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD, 'classifier': 'head.fc',
         **kwargs
     }
 
@@ -422,7 +475,8 @@ def _cfg(url='', **kwargs):
     'mambaout_base': _cfg(
         url='https://github.com/yuweihao/MambaOut/releases/download/model/mambaout_base.pth'),
     'mambaout_small_rw': _cfg(),
-    'mambaout_base_rw': _cfg(),
+    'mambaout_base_slim_rw': _cfg(),
+    'mambaout_base_plus_rw': _cfg(),
 }
 
 
@@ -480,12 +534,29 @@ def mambaout_small_rw(pretrained=False, **kwargs):
 
 
 @register_model
-def mambaout_base_rw(pretrained=False, **kwargs):
+def mambaout_base_slim_rw(pretrained=False, **kwargs):
     model_args = dict(
         depths=(3, 4, 27, 3),
         dims=(128, 256, 512, 768),
+        expansion_ratio=2.5,
+        conv_ratio=1.25,
         stem_mid_norm=False,
+        downsample='conv_nf',
+        ls_init_value=1e-6,
+        head_fn='norm_mlp',
+    )
+    return _create_mambaout('mambaout_base_slim_rw', pretrained=pretrained, **dict(model_args, **kwargs))
+
+
+@register_model
+def mambaout_base_plus_rw(pretrained=False, **kwargs):
+    model_args = dict(
+        depths=(3, 4, 27, 3),
+        dims=(128, 256, 512, 768),
+        expansion_ratio=3.0,
+        stem_mid_norm=False,
+        downsample='conv_nf',
         ls_init_value=1e-6,
         head_fn='norm_mlp',
     )
-    return _create_mambaout('mambaout_base_rw', pretrained=pretrained, **dict(model_args, **kwargs))
+    return _create_mambaout('mambaout_base_plus_rw', pretrained=pretrained, **dict(model_args, **kwargs))