[layers] fixed the concatenation of the conditional variable

bob/learn/tensorflow/layers/ConditionConcat.py

@@ -46,7 +46,5 @@ class ConditionConcat(Layer):
 
   def get_graph(self, y):
     
-    batch_size = input_layer.get_shape()[0]
-    yb = tf.reshape(y, [batch_size, 1, 1, self.conditional_dim])
-    return tf.concat([input_layer, y], 1)
+    return tf.concat([self.input_layer, y], 1)
 

bob/learn/tensorflow/layers/ImToCondFeatureMap.py

@@ -46,6 +46,8 @@ class ImToCondFeatureMap(Layer):
                                              use_gpu=False
                                             )
 
+    self.conditional_dim = conditional_dim
+    
     logger.info("+ adding a Concatenation layer ({0}) +".format(name))
     logger.info("\t conditional dimension = {0}".format(conditional_dim))
 
@@ -61,9 +63,11 @@ class ImToCondFeatureMap(Layer):
     name: y
     The conditioning vector (one-hot encoded)
 
-    """ 
-    # concatenate them to the image
-    x_shapes = input_layer.get_shape()
-    y_shapes = y.get_shape()
-    return tf.concat([input_layer, y*tf.ones([x_shapes[0], x_shapes[1], x_shapes[2], y_shapes[3]])], 3)
+    """
+    # concatenate to the image
+    batch_size = tf.shape(self.input_layer)[0]
+    yb = tf.reshape(y, [batch_size, 1, 1, self.conditional_dim])
+    x_shapes = self.input_layer.get_shape()
+    y_shapes = yb.get_shape()
+    return tf.concat([self.input_layer, yb*tf.ones([x_shapes[0], x_shapes[1], x_shapes[2], y_shapes[3]])], 3)