Optimize speed of CNBlock.

Author: datumbox

torchvision/prototype/models/convnext.py

@@ -37,29 +37,35 @@ def forward(self, x: Tensor) -> Tensor:
         return x
 
 
+class Permute(nn.Module):
+    def __init__(self, dims: List[int]):
+        super().__init__()
+        self.dims = dims
+
+    def forward(self, x):
+        return torch.permute(x, self.dims)
+
+
 class CNBlock(nn.Module):
     def __init__(
-        self, dim, layer_scale: float, stochastic_depth_prob: float, norm_layer: Callable[..., nn.Module]
+        self,
+        dim,
+        layer_scale: float,
+        stochastic_depth_prob: float,
+        norm_layer: Optional[Callable[..., nn.Module]] = None,
     ) -> None:
         super().__init__()
+        if norm_layer is None:
+            norm_layer = partial(nn.LayerNorm, eps=1e-6)
+
         self.block = nn.Sequential(
-            ConvNormActivation(
-                dim,
-                dim,
-                kernel_size=7,
-                groups=dim,
-                norm_layer=norm_layer,
-                activation_layer=None,
-                bias=True,
-            ),
-            ConvNormActivation(dim, 4 * dim, kernel_size=1, norm_layer=None, activation_layer=nn.GELU, inplace=None),
-            ConvNormActivation(
-                4 * dim,
-                dim,
-                kernel_size=1,
-                norm_layer=None,
-                activation_layer=None,
-            ),
+            nn.Conv2d(dim, dim, kernel_size=7, padding=3, groups=dim, bias=True),
+            Permute([0, 2, 3, 1]),
+            norm_layer(dim),
+            nn.Linear(in_features=dim, out_features=4 * dim, bias=True),
+            nn.GELU(),
+            nn.Linear(in_features=4 * dim, out_features=dim, bias=True),
+            Permute([0, 3, 1, 2]),
         )
         self.layer_scale = nn.Parameter(torch.ones(dim, 1, 1) * layer_scale)
         self.stochastic_depth = StochasticDepth(stochastic_depth_prob, "row")
@@ -142,7 +148,7 @@ def __init__(
             for _ in range(cnf.num_layers):
                 # adjust stochastic depth probability based on the depth of the stage block
                 sd_prob = stochastic_depth_prob * stage_block_id / (total_stage_blocks - 1.0)
-                stage.append(block(cnf.input_channels, layer_scale, sd_prob, norm_layer))
+                stage.append(block(cnf.input_channels, layer_scale, sd_prob))
                 stage_block_id += 1
             layers.append(nn.Sequential(*stage))
             if cnf.out_channels is not None:
@@ -213,7 +219,7 @@ def _convnext(
 
 class ConvNeXt_Tiny_Weights(WeightsEnum):
     IMAGENET1K_V1 = Weights(
-        url="https://download.pytorch.org/models/convnext_tiny-47b116bd.pth",
+        url="https://download.pytorch.org/models/convnext_tiny-47b116bd.pth",  # TODO: repackage
         transforms=partial(ImageNetEval, crop_size=224, resize_size=236),
         meta={
             **_COMMON_META,
@@ -227,7 +233,7 @@ class ConvNeXt_Tiny_Weights(WeightsEnum):
 
 class ConvNeXt_Small_Weights(WeightsEnum):
     IMAGENET1K_V1 = Weights(
-        url="https://download.pytorch.org/models/convnext_small-9aa23d28.pth",
+        url="https://download.pytorch.org/models/convnext_small-9aa23d28.pth",  # TODO: repackage
         transforms=partial(ImageNetEval, crop_size=224, resize_size=230),
         meta={
             **_COMMON_META,
@@ -241,7 +247,7 @@ class ConvNeXt_Small_Weights(WeightsEnum):
 
 class ConvNeXt_Base_Weights(WeightsEnum):
     IMAGENET1K_V1 = Weights(
-        url="https://download.pytorch.org/models/convnext_base-3b9f985d.pth",
+        url="https://download.pytorch.org/models/convnext_base-3b9f985d.pth",  # TODO: repackage
         transforms=partial(ImageNetEval, crop_size=224, resize_size=232),
         meta={
             **_COMMON_META,
@@ -255,7 +261,7 @@ class ConvNeXt_Base_Weights(WeightsEnum):
 
 class ConvNeXt_Large_Weights(WeightsEnum):
     IMAGENET1K_V1 = Weights(
-        url="https://download.pytorch.org/models/convnext_large-d73f62ac.pth",
+        url="https://download.pytorch.org/models/convnext_large-d73f62ac.pth",  # TODO: repackage
         transforms=partial(ImageNetEval, crop_size=224, resize_size=232),
         meta={
             **_COMMON_META,