Why split in 2 instead of 3 in gen_conv ?

[lchauvin]

Hello,

In v2, before the input of the coarse generator, you concatenate 3 images on the depth channel:

x = tf.concat([x, ones_x, ones_x*mask], axis=3)

but in gen_conv() you only split it in half:

x, y = tf.split(x, 2, 3)

Why is that ?