fix Pool2D output_size calculation to work with unknown (None) image …

…dimensions.

opendeep/models/utils/pooling.py

@@ -22,15 +22,29 @@ def _pool_out_size(imgshape, ds, st, padding, ignore_border=True):
     assert len(ds) == len(st), "stride and size need to have the same number of dimensions"
     ndims = len(ds)
     pooldims = list(imgshape[-ndims:])
-    pooldims = [dim + pad*2 for dim, pad in zip(pooldims, padding)]
+    pooldims = [dim + pad*2 if dim is not None and pad is not None
+                else None
+                for dim, pad in zip(pooldims, padding)]
 
     if ignore_border:
-        outdims = [(dim - size) // stride + 1 for dim, size, stride in zip(pooldims, ds, st)]
-        outdims = [max(outdim, 0) for outdim in outdims]
-    else:
-        outdims = [(dim - 1) // stride + 1 if stride >= size
-                   else max(0, (dim - 1 - size) // stride + 1) + 1
+        outdims = [(dim - size) // stride + 1
+                   if dim is not None and size is not None and stride is not None
+                   else None
                    for dim, size, stride in zip(pooldims, ds, st)]
+        outdims = [max(outdim, 0)
+                   if outdim is not None
+                   else None
+                   for outdim in outdims]
+    else:
+        outdims = []
+        for dim, size, stride in zip(pooldims, ds, st):
+            if dim is not None and size is not None and stride is not None:
+                if stride >= size:
+                    outdims.append((dim - 1) // stride + 1)
+                else:
+                    outdims.append(max(0, (dim - 1 - size) // stride + 1) + 1)
+            else:
+                outdims.append(None)
 
     rval = list(imgshape[:-ndims]) + outdims
     return rval