[Fix] Fix the output position of Swin-Transformer. (#947)

* [Fix] Fix the output position of Swin-Transformer.

* Rename `downsample` argument to `do_downsample`.

mmcls/models/backbones/swin_transformer.py

@@ -183,11 +183,11 @@ class SwinBlockSequence(BaseModule):
         else:
             self.downsample = None
 
-    def forward(self, x, in_shape):
+    def forward(self, x, in_shape, do_downsample=True):
         for block in self.blocks:
             x = block(x, in_shape)
 
-        if self.downsample:
+        if self.downsample is not None and do_downsample:
             x, out_shape = self.downsample(x, in_shape)
         else:
             out_shape = in_shape
@@ -232,6 +232,8 @@ class SwinTransformer(BaseBackbone):
         window_size (int): The height and width of the window. Defaults to 7.
         drop_rate (float): Dropout rate after embedding. Defaults to 0.
         drop_path_rate (float): Stochastic depth rate. Defaults to 0.1.
+        out_after_downsample (bool): Whether to output the feature map of a
+            stage after the following downsample layer. Defaults to False.
         use_abs_pos_embed (bool): If True, add absolute position embedding to
             the patch embedding. Defaults to False.
         interpolate_mode (str): Select the interpolate mode for absolute
@@ -301,6 +303,7 @@ class SwinTransformer(BaseBackbone):
                  drop_rate=0.,
                  drop_path_rate=0.1,
                  out_indices=(3, ),
+                 out_after_downsample=False,
                  use_abs_pos_embed=False,
                  interpolate_mode='bicubic',
                  with_cp=False,
@@ -329,6 +332,7 @@ class SwinTransformer(BaseBackbone):
         self.num_heads = self.arch_settings['num_heads']
         self.num_layers = len(self.depths)
         self.out_indices = out_indices
+        self.out_after_downsample = out_after_downsample
         self.use_abs_pos_embed = use_abs_pos_embed
         self.interpolate_mode = interpolate_mode
         self.frozen_stages = frozen_stages
@@ -392,9 +396,15 @@ class SwinTransformer(BaseBackbone):
             dpr = dpr[depth:]
             embed_dims.append(stage.out_channels)
 
+        if self.out_after_downsample:
+            self.num_features = embed_dims[1:]
+        else:
+            self.num_features = embed_dims[:-1]
+
         for i in out_indices:
             if norm_cfg is not None:
-                norm_layer = build_norm_layer(norm_cfg, embed_dims[i + 1])[1]
+                norm_layer = build_norm_layer(norm_cfg,
+                                              self.num_features[i])[1]
             else:
                 norm_layer = nn.Identity()
 
@@ -421,14 +431,17 @@ class SwinTransformer(BaseBackbone):
 
         outs = []
         for i, stage in enumerate(self.stages):
-            x, hw_shape = stage(x, hw_shape)
+            x, hw_shape = stage(
+                x, hw_shape, do_downsample=self.out_after_downsample)
             if i in self.out_indices:
                 norm_layer = getattr(self, f'norm{i}')
                 out = norm_layer(x)
                 out = out.view(-1, *hw_shape,
-                               stage.out_channels).permute(0, 3, 1,
+                               self.num_features[i]).permute(0, 3, 1,
                                                              2).contiguous()
                 outs.append(out)
+            if stage.downsample is not None and not self.out_after_downsample:
+                x, hw_shape = stage.downsample(x, hw_shape)
 
         return tuple(outs)
 

tests/test_models/test_backbones/test_swin_transformer.py

@@ -167,7 +167,7 @@ class TestSwinTransformer(TestCase):
         outs = model(imgs)
         self.assertIsInstance(outs, tuple)
         self.assertEqual(len(outs), 4)
-        for stride, out in zip([2, 4, 8, 8], outs):
+        for stride, out in zip([1, 2, 4, 8], outs):
             self.assertEqual(out.shape,
                              (1, 128 * stride, 56 // stride, 56 // stride))