use GRUCell instead of GRU

lucidrains · Jul 23, 2020 · cb3a182 · cb3a182
1 parent 272cab9
commit cb3a182
Show file tree

Hide file tree

Showing 3 changed files with 9 additions and 9 deletions.
diff --git a/setup.py b/setup.py
@@ -3,7 +3,7 @@
 setup(
   name = 'slot_attention',
   packages = find_packages(),
-  version = '0.0.6',
+  version = '0.1.0',
   license='MIT',
   description = 'Implementation of Slot Attention in Pytorch',
   author = 'Phil Wang',

diff --git a/slot_attention/slot_attention.py b/slot_attention/slot_attention.py
@@ -16,7 +16,7 @@ def __init__(self, num_slots, dim, iters = 3, eps = 1e-8, hidden_dim = 128):
         self.to_k = nn.Linear(dim, dim, bias = False)
         self.to_v = nn.Linear(dim, dim, bias = False)
 
-        self.gru = nn.GRU(dim, dim)
+        self.gru = nn.GRUCell(dim, dim)
 
         hidden_dim = max(dim, hidden_dim)
 
@@ -53,9 +53,9 @@ def forward(self, inputs, num_slots = None):
 
             updates = torch.einsum('bjd,bij->bid', v, attn)
 
-            slots, _ = self.gru(
-                updates.reshape(1, -1, d),
-                slots_prev.reshape(1, -1, d)
+            slots = self.gru(
+                updates.reshape(-1, d),
+                slots_prev.reshape(-1, d)
             )
 
             slots = slots.reshape(b, -1, d)

diff --git a/slot_attention/slot_attention_experimental.py b/slot_attention/slot_attention_experimental.py
@@ -42,17 +42,17 @@ def forward(self, x):
 class GatedResidual(nn.Module):
     def __init__(self, dim, fn):
         super().__init__()
-        self.gru = nn.GRU(dim, dim)
+        self.gru = nn.GRUCell(dim, dim)
         self.fn = fn
     def forward(self, *args):
         inputs = args[0]
         b, _, d = inputs.shape
 
         updates = self.fn(*args)
 
-        inputs, _ = self.gru(
-            updates.reshape(1, -1, d),
-            inputs.reshape(1, -1, d)
+        inputs = self.gru(
+            updates.reshape(-1, d),
+            inputs.reshape(-1, d)
         )
         return inputs.reshape(b, -1, d)