Update.

[pytorch.git] / ae_size.py

diff --git a/ae_size.py b/ae_size.py
index 7bef9f5..8afb101 100755 (executable)
--- a/ae_size.py
+++ b/ae_size.py
@@ -11,27 +11,34 @@ def minimal_input_size(w, layer_specs):
     if layer_specs == []:
         return w
     else:
-        k, s = layer_specs[0]
-        w = math.ceil((w - k) / s) + 1
-        w = minimal_input_size(w, layer_specs[1:])
-        return int((w - 1) * s + k)
+        kernel_size, stride = layer_specs[0]
+        v = int(math.ceil((w - kernel_size) / stride)) + 1
+        v = minimal_input_size(v, layer_specs[1:])
+        return (v - 1) * stride + kernel_size
 
 ######################################################################
 
-layer_specs = [ (11, 5), (5, 2), (3, 2), (3, 2) ]
+# Dummy test
 
-layers = []
-for l in layer_specs:
-    layers.append(nn.Conv2d(1, 1, l[0], l[1]))
+if __name__ == "__main__":
 
-for l in reversed(layer_specs):
-    layers.append(nn.ConvTranspose2d(1, 1, l[0], l[1]))
+    layer_specs = [ (17, 5), (5, 4), (3, 2), (3, 2) ]
 
-m = nn.Sequential(*layers)
+    layers = []
 
-h = minimal_input_size(240, layer_specs)
-w = minimal_input_size(320, layer_specs)
+    for kernel_size, stride in layer_specs:
+        layers.append(nn.Conv2d(1, 1, kernel_size, stride))
 
-x = Tensor(1, 1, h, w).normal_()
+    for kernel_size, stride in reversed(layer_specs):
+        layers.append(nn.ConvTranspose2d(1, 1, kernel_size, stride))
 
-print(x.size(), m(x).size())
+    m = nn.Sequential(*layers)
+
+    h = minimal_input_size(240, layer_specs)
+    w = minimal_input_size(320, layer_specs)
+
+    x = Tensor(1, 1, h, w).normal_()
+
+    print(x.size(), m(x).size())
+
+######################################################################