Bit more Hiera fiddling

timm/models/hiera.py

@@ -47,7 +47,7 @@ def conv_nd(n: int) -> Type[nn.Module]:
     return [nn.Identity, nn.Conv1d, nn.Conv2d, nn.Conv3d][n]
 
 
-def get_resized_mask(target_size: torch.Size, mask: torch.Tensor) -> torch.Tensor:
+def get_resized_mask(target_size: List[int], mask: torch.Tensor) -> torch.Tensor:
     # target_size: [(T), (H), W]
     # (spatial) mask: [B, C, (t), (h), w]
     if mask is None:
@@ -59,23 +59,6 @@ def get_resized_mask(target_size: torch.Size, mask: torch.Tensor) -> torch.Tenso
     return mask
 
 
-def do_masked_conv(
-        x: torch.Tensor,
-        conv: nn.Module,
-        mask: Optional[torch.Tensor] = None,
-) -> torch.Tensor:
-    """Zero-out the masked regions of the input before conv.
-    Prevents leakage of masked regions when using overlapping kernels.
-    """
-    if conv is None:
-        return x
-    if mask is None:
-        return conv(x)
-
-    mask = get_resized_mask(target_size=x.shape[2:], mask=mask)
-    return conv(x * mask.bool())
-
-
 def undo_windowing(
         x: torch.Tensor,
         shape: List[int],
@@ -145,7 +128,6 @@ class Unroll(nn.Module):
         Output: Patch embeddings [B, N, C] permuted such that [B, 4, N//4, C].max(1) etc. performs MaxPoolNd
         """
         B, _, C = x.shape
-
         cur_size = self.size
         x = x.view(*([B] + cur_size + [C]))
 
@@ -332,6 +314,7 @@ class HieraBlock(nn.Module):
             act_layer: nn.Module = nn.GELU,
             q_stride: int = 1,
             window_size: int = 0,
+            use_expand_proj: bool = True,
             use_mask_unit_attn: bool = False,
     ):
         super().__init__()
@@ -341,8 +324,14 @@ class HieraBlock(nn.Module):
 
         self.norm1 = norm_layer(dim)
         if dim != dim_out:
-            self.proj = nn.Linear(dim, dim_out)
+            self.do_expand = True
+            if use_expand_proj:
+                self.proj = nn.Linear(dim, dim_out)
+            else:
+                assert dim_out == dim * 2
+                self.proj = None
         else:
+            self.do_expand = False
             self.proj = None
         self.attn = MaskUnitAttention(
             dim,
@@ -362,9 +351,17 @@ class HieraBlock(nn.Module):
     def forward(self, x: torch.Tensor) -> torch.Tensor:
         # Attention + Q Pooling
         x_norm = self.norm1(x)
-        if self.proj is not None:
+        if self.do_expand:
-            x = self.proj(x_norm)
+            if self.proj is not None:
-            x = x.view(x.shape[0], self.attn.q_stride, -1, x.shape[-1]).amax(dim=1)  # max-pool
+                x = self.proj(x_norm)
+                x = x.view(x.shape[0], self.attn.q_stride, -1, x.shape[-1]).amax(dim=1)  # max-pool
+            else:
+                x = torch.cat([
+                    x.view(x.shape[0], self.attn.q_stride, -1, x.shape[-1]).amax(dim=1),  # max-pool
+                    x.view(x.shape[0], self.attn.q_stride, -1, x.shape[-1]).mean(dim=1),  # avg-pool
+                    ],
+                    dim=-1,
+                )
         x = x + self.drop_path1(self.attn(x_norm))
 
         # MLP
@@ -419,7 +416,11 @@ class PatchEmbed(nn.Module):
             x: torch.Tensor,
             mask: Optional[torch.Tensor] = None,
     ) -> torch.Tensor:
-        x = do_masked_conv(x, self.proj, mask)
+        if mask is not None:
+            mask = get_resized_mask(target_size=x.shape[2:], mask=mask)
+            x = self.proj(x * mask.to(torch.bool))
+        else:
+            x = self.proj(x)
         if self.reshape:
             x = x.reshape(x.shape[0], x.shape[1], -1).transpose(2, 1)
         return x
@@ -570,10 +571,10 @@ class Hiera(nn.Module):
 
     @torch.jit.ignore
     def no_weight_decay(self):
-        if self.sep_pos_embed:
+        if self.pos_embed is not None:
-            return ["pos_embed_spatial", "pos_embed_temporal"]
-        else:
             return ["pos_embed"]
+        else:
+            return ["pos_embed_spatial", "pos_embed_temporal"]
 
     def get_random_mask(self, x: torch.Tensor, mask_ratio: float) -> torch.Tensor:
         """