经典神经网络架构参考 v1.0(2)

digraph AlexNetL678 {
  rankdir=BT
    node [
    style=filled, 
    color=Black
    fontcolor=White, 
    fillcolor="#30638e", 
    fontname="SimHei",
    fontsize=32,
    width=5, height=2,
    shape="box",
  ]
    featmap52 [label="[BatchSize,\nW=6, H=6, C=128x2]", shape="Mrecord"]
    reshape [label="reshape([4096])"]
    linear6 [label="Linear6\n[4096, 4096]"]
    relu6 [label="Relu"]
    dropout6 [label="Dropout"]
    linear7 [label="Linear7\n[4096, 4096]"]
    relu7 [label="Relu"]
    dropout7 [label="Dropout"]
    linear8 [label="Linear7\n[4096, 1000]"]
    softmax [label="Softmax"]
    oup [label="输出\n[BatchSize, 1000]", shape="Mrecord"]
    featmap52 -> reshape -> linear6 -> relu6 -> dropout6 ->
        linear7 -> relu7 -> dropout7 -> linear8 -> softmax -> oup
}

digraph VGG16 {
  rankdir=BT
    node [
    style=filled, 
    color=Black
    fontcolor=White, 
    fillcolor="#30638e", 
    fontname="SimHei",
    fontsize=32,
    width=5, height=2,
    shape="box",
  ]
    inp [label="输入\n[BatchSize,\nW=224, H=224, C=3]", shape="Mrecord"]
  convblock1 [label="ConvBlock1\n[In=3, Out=64]"]
  featmap1 [label="[BatchSize\nW=112, H=112, C=64]", shape="Mrecord"]
  convblock2 [label="ConvBlock2\n[In=64, Out=128]"]
  featmap2 [label="[BatchSize\nW=56, H=56, C=128]", shape="Mrecord"]
  convblock3 [label="ConvBlock3\n[In=128, Out=256]"]
  featmap3 [label="[BatchSize\nW=28, H=28, C=256]", shape="Mrecord"]
  convblock4 [label="ConvBlock4\n[In=256, Out=512]"]
  featmap4 [label="[BatchSize\nW=14, H=14, C=512]", shape="Mrecord"]
  convblock5 [label="ConvBlock5\n[In=512, Out=512]"]
  featmap5 [label="[BatchSize\nW=7, H=7, C=512]", shape="Mrecord"]
  reshape [label="reshape([7x7x512])"]
  linear1 [label="Linear1\n[7x7x512, 4096]"]
  relu1 [label="Relu"]
  linear2 [label="Linear2\n[4096, 4096]"]
  relu2 [label="Relu"]
  linear3 [label="Linear3\n[4096, 1000]"]
  relu3 [label="Relu"]
  linear4 [label="Linear4\n[1000, 1000]"]
  softmax [label="Softmax"]
  oup [label="输出\n[BatchSize, 1000]", shape="Mrecord"]
  inp -> convblock1 -> featmap1 -> convblock2 -> featmap2 ->
    convblock3 -> featmap3 -> convblock4 -> featmap4 -> 
    convblock5 -> featmap5 -> reshape -> linear1 -> relu1 ->
    linear2 -> relu2 -> linear3 -> relu3 -> linear4 -> softmax ->
    oup
}

digraph VGG16ConvBlock {
  rankdir=BT
    node [
    style=filled, 
    color=Black
    fontcolor=White, 
    fillcolor="#30638e", 
    fontname="SimHei",
    fontsize=32,
    width=5, height=2,
    shape="box",
  ]
    inp [label="输入\n[BatchSize, W=BlockW,\n H=BlockH, C=BlockIn]", shape="Mrecord"]
  conv1 [label="Conv2D#1\n[K=3, P=1,\n In=BlockIn,\n Out=BlockOut]"]
  conv2 [label="Conv2D#2\n[K=3, P=1,\n In=BlockOut,\n Out=BlockOut]"]
  relu  [label="relu"]
  maxpool [label="MaxPool2D\n[K=2, S=2]"]
    oup [label="输出\n[BatchSize,\n W=BlockW/2,\n H=BlockH/2,\n C=BlockOut]", shape="Mrecord"]
  inp -> conv1 -> conv2 -> relu -> maxpool -> oup
}

digraph ResNet18 {
  rankdir=BT
    node [
    style=filled, 
    color=Black
    fontcolor=White, 
    fillcolor="#30638e", 
    fontname="SimHei",
    fontsize=32,
    width=5, height=2,
    shape="box",
  ]
    inp [label="输入\n[BatchSize,\nW=224, H=224, C=3]", shape="Mrecord"]
    conv0 [label="Conv2D#0\n[K=7, S=2, P=3, In=3, Out=64]"]
    maxpool0 [label="MaxPool2D#0\n[K=3, S=2, P=1]"]
    featmap0 [label="[BatchSize,\nW=56, H=56, C=64]", shape="Mrecord"]
  convblock1 [label="ConvBlock1\n[C=64]"]
  convblock2 [label="ConvBlock2\n[C=64]"]
    featmap2 [label="[BatchSize,\nW=56, H=56, C=64]", shape="Mrecord"]
    downconvblock3 [label="DownConvBlock3\n[In=64, Out=128]"]
  convblock4 [label="ConvBlock4\n[C=128]"]
    featmap4 [label="[BatchSize,\nW=28, H=28, C=128]", shape="Mrecord"]
    downconvblock5 [label="DownConvBlock5\n[In=128, Out=256]"]
  convblock6 [label="ConvBlock6\n[C=256]"]
    featmap6 [label="[BatchSize,\nW=14, H=14, C=256]", shape="Mrecord"]
    downconvblock7 [label="DownConvBlock7\n[In=256, Out=512]"]
  convblock8 [label="ConvBlock8\n[C=512]"]
    featmap8 [label="[BatchSize,\nW=7, H=7, C=512]", shape="Mrecord"]
    avgpool [label="AvgPool2D\n[K=7, S=7]"]
    featmap9 [label="[BatchSize,\nW=1, H=1, C=512]", shape="Mrecord"]
    reshape [label="reshape([512])"]
    linear [label="Linear\n[512, 1000]"]
    softmax [label="Softmax"]
  oup [label="输出\n[BatchSize, 1000]", shape="Mrecord"]
    inp -> conv0 -> maxpool0 -> featmap0 -> convblock1 -> convblock2 -> 
        featmap2 -> downconvblock3 -> convblock4 -> featmap4 ->
        downconvblock5 -> convblock6 -> featmap6 ->
        downconvblock7 -> convblock8 -> featmap8 -> avgpool ->
        featmap9 -> reshape -> linear -> softmax -> oup
}

digraph ResNet18ConvBlock {
  rankdir=BT
    node [
    style=filled, 
    color=Black
    fontcolor=White, 
    fillcolor="#30638e", 
    fontname="SimHei",
    fontsize=32,
    width=5, height=2,
    shape="box",
  ]
    
    inp [label="输入\n[BatchSize,\nW=BlockW, \nH=BlockH, C=BlockC]", shape="Mrecord"]
    conv1 [label="Conv2D#1\n[K=3, P=1,\n In=BlockC,\n Out=BlockC]"]
    relu1 [label="Relu"]
    conv2 [label="Conv2D#2\n[K=3, P=1,\n In=BlockC,\n Out=BlockC]"]
    add   [label="+"]
    relu2 [label="Relu"]
    inp -> conv1 -> relu1 -> conv2 -> add -> relu2 -> oup
    inp -> add
}