Update davit.py

timm/models/davit.py

@@ -23,8 +23,9 @@ import torch.nn as nn
 import torch.nn.functional as F
 from torch import Tensor
 from .helpers import build_model_with_cfg
-from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
+from .data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
 from .layers import DropPath, to_2tuple, trunc_normal_, SelectAdaptivePool2d, ClassifierHead, Mlp
+from .features import FeatureInfo
 from collections import OrderedDict
 import torch.utils.checkpoint as checkpoint
 from .pretrained import generate_default_cfgs
@@ -381,8 +382,7 @@ class DaViT(nn.Module):
         drop_rate=0.,
         attn_drop_rate=0.,
         num_classes=1000,
-        global_pool='avg',
-        **kwargs
+        global_pool='avg'
     ):
         super().__init__()
 
@@ -439,12 +439,7 @@ class DaViT(nn.Module):
             ])
             
             self.stages.add_module(f'stage_{stage_id}', stage)
-            
-            self.feature_info += [dict(
-                num_chs=self.embed_dims[stage_id],
-                reduction = 2,
-                module=f'stages.stage_{stage_id}.{depths[stage_id] - 1}.{len(attention_types) - 1}.mlp')]
-
+            self.feature_info += [dict(num_chs=self.embed_dims[item], reduction=2, module=f'stages.stage_{stage_id}')]
 
         self.norms = norm_layer(self.num_features)
         self.head = ClassifierHead(self.num_features, num_classes, pool_type=global_pool, drop_rate=drop_rate)
@@ -498,15 +493,7 @@ class DaViT(nn.Module):
         # non-normalized pyramid features + corresponding sizes
         return features, sizes
     
-    def forward_pyramid_features(self, x):
-        x, sizes = self.forward_network(x)
-        outs = []
-        for i, out in enumerate(x):
-            H, W = sizes[i]
-            outs.append(out.view(-1, H, W, self.embed_dims[i]).permute(0, 3, 1, 2).contiguous())
-        
-        return outs
-    
+
     def forward_features(self, x):
         x, sizes = self.forward_network(x)
         # take final feature and norm
@@ -523,6 +510,22 @@ class DaViT(nn.Module):
         x = self.forward_head(x)
         return x
 
+class DaViTFeatures(DaViT):
+
+    def __init__(*args):
+        super(DaViT, self).__init__(*args, **kwargs)
+        default_out_indices = tuple(i for i, _ in enumerate(kwargs.get('depths', (1, 1, 3, 1))))
+        out_indices = kwargs.pop('out_indices', default_out_indices)
+        self.feature_info = FeatureInfo(self.feature_info, out_indices)
+    
+    def forward_pyramid_features(self, x) -> List[Tensor]:
+        x, sizes = self.forward_network(x)
+        outs = []
+        for i, out in enumerate(x):
+            H, W = sizes[i]
+            outs.append(out.view(-1, H, W, self.embed_dims[i]).permute(0, 3, 1, 2).contiguous())
+        
+        return outs
     
 
 def checkpoint_filter_fn(state_dict, model):
@@ -543,15 +546,23 @@ def checkpoint_filter_fn(state_dict, model):
     
     
 def _create_davit(variant, pretrained=False, **kwargs):
-    default_out_indices = tuple(i for i, _ in enumerate(kwargs.get('depths', (1, 1, 3, 1))))
-    out_indices = kwargs.pop('out_indices', default_out_indices)
+    model_cls = HighResolutionNet
+    features_only = False
+    kwargs_filter = None
+    if model_kwargs.pop('features_only', False):
+        model_cls = HighResolutionNetFeatures
+        kwargs_filter = ('num_classes', 'global_pool')
+        features_only = True
     model = build_model_with_cfg(
-        DaViT,
+        model_cls,
         variant,
         pretrained,
         pretrained_filter_fn=checkpoint_filter_fn,
         feature_cfg=dict(flatten_sequential=True, out_indices=out_indices),
         **kwargs)
+    if features_only:
+        model.pretrained_cfg = pretrained_cfg_for_features(model.default_cfg)
+        model.default_cfg = model.pretrained_cfg  # backwards compat
     return model