Encoder: VGG16 Res50

Decoder: FCN32s FCN16s FCN8s PSPNet Deeplab v2

TODO

1. get the edge as the soft gate.
2. get the stage-wise heatmap as the soft gate.

detection great thanks to https://itnext.io/implementing-yolo-v3-in-tensorflow-tf-slim-c3c55ff59dbe

ResNet101: 49364+3*(25664+96464+64256)+4*(512128+9128128+128512)+23*(1024256+9256256+2561024)+3*(2048512+9512512+5122048)=40326336 VGG16: 9*(364+6464+64+128+128128+128256+2256256+256512+5512*512)=14638464

Simple=49*(364+76464)=1414336 Simple_Conv_Part = 49(364+36464)=611520 Deconv = 4946464=802816 Simple_Stack_Conv_Part = 27*(364+364*64)=336960