Upgraded to TF version 1.13.2

Removed the wait at first launch for most graphics cards.

Increased speed of training by 10-20%, but you have to retrain all models from scratch.

SAEHD:

added option 'use float16'
	Experimental option. Reduces the model size by half.
	Increases the speed of training.
	Decreases the accuracy of the model.
	The model may collapse or not train.
	Model may not learn the mask in large resolutions.

true_face_training option is replaced by
"True face power". 0.0000 .. 1.0
Experimental option. Discriminates the result face to be more like the src face. Higher value - stronger discrimination.
Comparison - https://i.imgur.com/czScS9q.png

core/leras/initializers.py

@@ -11,17 +11,14 @@ def initialize_initializers(nn):
 
     class initializers():
         class ca (init_ops.Initializer):
-            def __init__(self, dtype=None):
-                pass
-
             def __call__(self, shape, dtype=None, partition_info=None):
-                return tf.zeros( shape, name="_cai_")
+                return tf.zeros( shape, dtype=dtype, name="_cai_")
 
             @staticmethod
             def generate_batch( data_list, eps_std=0.05 ):
                 # list of (shape, np.dtype)
                 return CAInitializerSubprocessor (data_list).run()
-  
+
     nn.initializers = initializers
 
 class CAInitializerSubprocessor(Subprocessor):
@@ -62,7 +59,7 @@ class CAInitializerSubprocessor(Subprocessor):
         x = x * np.sqrt( (2/fan_in) / np.var(x) )
         x = np.transpose( x, (2, 3, 1, 0) )
         return x.astype(dtype)
-                
+
     class Cli(Subprocessor.Cli):
         #override
         def process_data(self, data):