|
|
|
@ -438,10 +438,12 @@ class DaViTStage(nn.Module):
|
|
|
|
|
x, size = self.patch_embed(x, size)
|
|
|
|
|
x : Tensor = x
|
|
|
|
|
size : Tuple[int, int] = size
|
|
|
|
|
dim1 : int = size[0]
|
|
|
|
|
dim2 : int = size[1]
|
|
|
|
|
if self.grad_checkpointing and not torch.jit.is_scripting():
|
|
|
|
|
x, size = checkpoint_seq(self.blocks, x, size)
|
|
|
|
|
x, size = checkpoint_seq(self.blocks, x, (dim1, dim2))
|
|
|
|
|
else:
|
|
|
|
|
x, size = self.blocks(x, size)
|
|
|
|
|
x, size = self.blocks(x, (dim1, dim2))
|
|
|
|
|
|
|
|
|
|
x : Tensor = x
|
|
|
|
|
size : Tuple[int, int] = size
|
|
|
|
@ -592,7 +594,7 @@ class DaViT(nn.Module):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def forward_features(self, x):
|
|
|
|
|
def forward_features(self, x : Tensor):
|
|
|
|
|
#x, sizes = self.forward_network(x)
|
|
|
|
|
size: Tuple[int, int] = (x.size(2), x.size(3))
|
|
|
|
|
x, size = self.stages(x, size)
|
|
|
|
|