upd docs

Author: cirobr

README.md

@@ -64,6 +64,8 @@ If ch_out > 1, output mask activation becomes softmax. For instance, a model wit
 
 UNet5() has internally five encoder/decoder stages, each of them delivering features with respectivelly $[64, 128, 256, 512, 1024]$ channels.
 
+UNet4() has four encoder/decoder stages and $[64, 128, 256, 512]$ channels.
+
 Argument $alpha$ modulates the number of internal channels proportionally. For instance, $alpha == 2$ delivers $[32, 64, 128, 256, 512]$ channels.
 
 Argument $verbose == false$ delivers output mask with same frame size as input images.