Control first downsample stride in ResNet (#26374)

src/transformers/models/resnet/configuration_resnet.py

@@ -59,6 +59,8 @@ class ResNetConfig(BackboneConfigMixin, PretrainedConfig):
             are supported.
         downsample_in_first_stage (`bool`, *optional*, defaults to `False`):
             If `True`, the first stage will downsample the inputs using a `stride` of 2.
+        downsample_in_bottleneck (`bool`, *optional*, defaults to `False`):
+            If `True`, the first conv 1x1 in ResNetBottleNeckLayer will downsample the inputs using a `stride` of 2.
         out_features (`List[str]`, *optional*):
             If used as backbone, list of features to output. Can be any of `"stem"`, `"stage1"`, `"stage2"`, etc.
             (depending on how many stages the model has). If unset and `out_indices` is set, will default to the
@@ -94,6 +96,7 @@ def __init__(
         layer_type="bottleneck",
         hidden_act="relu",
         downsample_in_first_stage=False,
+        downsample_in_bottleneck=False,
         out_features=None,
         out_indices=None,
         **kwargs,
@@ -108,6 +111,7 @@ def __init__(
         self.layer_type = layer_type
         self.hidden_act = hidden_act
         self.downsample_in_first_stage = downsample_in_first_stage
+        self.downsample_in_bottleneck = downsample_in_bottleneck
         self.stage_names = ["stem"] + [f"stage{idx}" for idx in range(1, len(depths) + 1)]
         self._out_features, self._out_indices = get_aligned_output_features_output_indices(
             out_features=out_features, out_indices=out_indices, stage_names=self.stage_names

src/transformers/models/resnet/modeling_resnet.py

@@ -149,11 +149,18 @@ class ResNetBottleNeckLayer(nn.Module):
     A classic ResNet's bottleneck layer composed by three `3x3` convolutions.
 
     The first `1x1` convolution reduces the input by a factor of `reduction` in order to make the second `3x3`
-    convolution faster. The last `1x1` convolution remaps the reduced features to `out_channels`.
+    convolution faster. The last `1x1` convolution remaps the reduced features to `out_channels`. If
+    `downsample_in_bottleneck` is true, downsample will be in the first layer instead of the second layer.
     """
 
     def __init__(
-        self, in_channels: int, out_channels: int, stride: int = 1, activation: str = "relu", reduction: int = 4
+        self,
+        in_channels: int,
+        out_channels: int,
+        stride: int = 1,
+        activation: str = "relu",
+        reduction: int = 4,
+        downsample_in_bottleneck: bool = False,
     ):
         super().__init__()
         should_apply_shortcut = in_channels != out_channels or stride != 1
@@ -162,8 +169,10 @@ def __init__(
             ResNetShortCut(in_channels, out_channels, stride=stride) if should_apply_shortcut else nn.Identity()
         )
         self.layer = nn.Sequential(
-            ResNetConvLayer(in_channels, reduces_channels, kernel_size=1),
-            ResNetConvLayer(reduces_channels, reduces_channels, stride=stride),
+            ResNetConvLayer(
+                in_channels, reduces_channels, kernel_size=1, stride=stride if downsample_in_bottleneck else 1
+            ),
+            ResNetConvLayer(reduces_channels, reduces_channels, stride=stride if not downsample_in_bottleneck else 1),
             ResNetConvLayer(reduces_channels, out_channels, kernel_size=1, activation=None),
         )
         self.activation = ACT2FN[activation]
@@ -194,10 +203,18 @@ def __init__(
 
         layer = ResNetBottleNeckLayer if config.layer_type == "bottleneck" else ResNetBasicLayer
 
+        if config.layer_type == "bottleneck":
+            first_layer = layer(
+                in_channels,
+                out_channels,
+                stride=stride,
+                activation=config.hidden_act,
+                downsample_in_bottleneck=config.downsample_in_bottleneck,
+            )
+        else:
+            first_layer = layer(in_channels, out_channels, stride=stride, activation=config.hidden_act)
         self.layers = nn.Sequential(
-            # downsampling is done in the first layer with stride of 2
-            layer(in_channels, out_channels, stride=stride, activation=config.hidden_act),
-            *[layer(out_channels, out_channels, activation=config.hidden_act) for _ in range(depth - 1)],
+            first_layer, *[layer(out_channels, out_channels, activation=config.hidden_act) for _ in range(depth - 1)]
         )
 
     def forward(self, input: Tensor) -> Tensor: