Implementation of vanilla unet & unet with self-attention
In self-attention implementation conv3x3 blocks are replaced by conv1x1+self-attention+conv1x1. See unet_plus_attention/model/model.py & unet_plus_attention/model/unet_blocks_builder.py for more details