fix simsiam, thanks to @chingisooinar

byol_pytorch/byol_pytorch.py

@@ -75,25 +75,32 @@ def update_moving_average(ema_updater, ma_model, current_model):
 
 # MLP class for projector and predictor
 
-class MLP(nn.Module):
-    def __init__(self, dim, projection_size, hidden_size = 4096):
-        super().__init__()
-        self.net = nn.Sequential(
-            nn.Linear(dim, hidden_size),
-            nn.BatchNorm1d(hidden_size),
-            nn.ReLU(inplace=True),
-            nn.Linear(hidden_size, projection_size)
-        )
-
-    def forward(self, x):
-        return self.net(x)
+def MLP(dim, projection_size, hidden_size=4096):
+    return nn.Sequential(
+        nn.Linear(dim, hidden_size),
+        nn.BatchNorm1d(hidden_size),
+        nn.ReLU(inplace=True),
+        nn.Linear(hidden_size, projection_size)
+    )
+
+def SimSiamMLP(dim, projection_size, hidden_size=4096):
+    return nn.Sequential(
+        nn.Linear(dim, hidden_size, bias=False),
+        nn.BatchNorm1d(hidden_size),
+        nn.ReLU(inplace=True),
+        nn.Linear(hidden_size, hidden_size, bias=False),
+        nn.BatchNorm1d(hidden_size),
+        nn.ReLU(inplace=True),
+        nn.Linear(hidden_size, projection_size, bias=False),
+        nn.BatchNorm1d(projection_size, affine=False)
+    )
 
 # a wrapper class for the base neural network
 # will manage the interception of the hidden layer output
 # and pipe it into the projecter and predictor nets
 
 class NetWrapper(nn.Module):
-    def __init__(self, net, projection_size, projection_hidden_size, layer = -2):
+    def __init__(self, net, projection_size, projection_hidden_size, layer = -2, use_simsiam_mlp = False):
         super().__init__()
         self.net = net
         self.layer = layer
@@ -102,6 +109,8 @@ def __init__(self, net, projection_size, projection_hidden_size, layer = -2):
         self.projection_size = projection_size
         self.projection_hidden_size = projection_hidden_size
 
+        self.use_simsiam_mlp = use_simsiam_mlp
+
         self.hidden = {}
         self.hook_registered = False
 
@@ -127,7 +136,8 @@ def _register_hook(self):
     @singleton('projector')
     def _get_projector(self, hidden):
         _, dim = hidden.shape
-        projector = MLP(dim, self.projection_size, self.projection_hidden_size)
+        create_mlp_fn = MLP if not self.use_simsiam_mlp else SimSiamMLP
+        projector = create_mlp_fn(dim, self.projection_size, self.projection_hidden_size)
         return projector.to(hidden)
 
     def get_representation(self, x):
@@ -195,7 +205,7 @@ def __init__(
         self.augment1 = default(augment_fn, DEFAULT_AUG)
         self.augment2 = default(augment_fn2, self.augment1)
 
-        self.online_encoder = NetWrapper(net, projection_size, projection_hidden_size, layer=hidden_layer)
+        self.online_encoder = NetWrapper(net, projection_size, projection_hidden_size, layer=hidden_layer, use_simsiam_mlp=not use_momentum)
 
         self.use_momentum = use_momentum
         self.target_encoder = None

setup.py

@@ -3,13 +3,16 @@
 setup(
   name = 'byol-pytorch',
   packages = find_packages(exclude=['examples']),
-  version = '0.5.7',
+  version = '0.6.0',
   license='MIT',
   description = 'Self-supervised contrastive learning made simple',
   author = 'Phil Wang',
   author_email = '[email protected]',
   url = 'https://github.com/lucidrains/byol-pytorch',
-  keywords = ['self-supervised learning', 'artificial intelligence'],
+  keywords = [
+      'self-supervised learning',
+      'artificial intelligence'
+  ],
   install_requires=[
       'torch>=1.6',
       'torchvision>=0.8'