added untrained model Inception v3

src/VisionModel/VisionModel.jl

@@ -7,10 +7,11 @@ include("alexnet.jl")
 include("vgg.jl")
 include("resnet.jl")
 include("googlenet.jl")
+include("inception.jl")
 
 export  alexnet,
         vgg11, vgg11bn, vgg13, vgg13bn, vgg16, vgg16bn, vgg19, vgg19bn,
         ResNet18, ResNet34, ResNet50, ResNet101, ResNet152,
-        googlenet
+        googlenet, inception3
 
 end

src/VisionModel/inception.jl

@@ -0,0 +1,176 @@
+function conv_block(kernelsize::Tuple{Int64,Int64}, inplanes::Int64, outplanes::Int64; stride::Int64=1, pad::Union{Int64,Tuple{Int64,Int64}}=0)
+    conv_layer = []
+    push!(conv_layer, Conv(kernelsize, inplanes => outplanes, stride=stride, pad=pad))
+    push!(conv_layer, BatchNorm(outplanes, relu))
+    return conv_layer
+end
+
+function inception_a(inplanes, pool_proj)
+    branch1x1 = Chain(conv_block((1, 1), inplanes, 64)...)
+
+    branch5x5_1 = Chain(conv_block((1, 1), inplanes, 48)...)
+    branch5x5_2 = Chain(conv_block((5, 5), 48, 64; pad=2)...)
+
+    branch3x3dbl_1 = Chain(conv_block((1, 1), inplanes, 64)...)
+    branch3x3dbl_2 = Chain(conv_block((3, 3), 64, 96; pad=1)...)
+    branch3x3dbl_3 = Chain(conv_block((3, 3), 96, 96; pad=1)...)
+
+    branch_pool = Chain(MeanPool((3, 3), pad=1, stride=1),
+                        conv_block((1, 1), inplanes, pool_proj)...)
+
+    layer = x -> begin 
+        y1 = branch1x1(x)
+
+        y2 = branch5x5_1(x)
+        y2 = branch5x5_2(y2)
+
+        y3 = branch3x3dbl_1(x)
+        y3 = branch3x3dbl_2(y3)
+        y3 = branch3x3dbl_3(y3)
+
+        y4 = branch_pool(x)
+
+        return cat(y1, y2, y3, y4; dims=3)
+    end
+end
+
+function inception_b(inplanes)
+    branch3x3 = Chain(conv_block((3, 3), inplanes, 384; stride=2)...)
+
+    branch3x3dbl_1 = Chain(conv_block((1, 1), inplanes, 64)...)
+    branch3x3dbl_2 = Chain(conv_block((3, 3), 64, 96; pad=1)...)
+    branch3x3dbl_3 = Chain(conv_block((3, 3), 96, 96; stride=2)...)
+
+    branch_pool = Chain(MaxPool((3, 3), stride=2))
+
+    layer = x -> begin
+        y1 = branch3x3(x)
+
+        y2 = branch3x3dbl_1(x)
+        y2 = branch3x3dbl_2(y2)
+        y2 = branch3x3dbl_3(y2)
+
+        y3 = branch_pool(x)
+
+        return cat(y1, y2, y3; dims=3)
+    end
+end
+
+function inception_c(inplanes, c7)
+    branch1x1 = Chain(conv_block((1, 1), inplanes, 192)...)
+
+    branch7x7_1 = Chain(conv_block((1, 1), inplanes, c7)...)
+    branch7x7_2 = Chain(conv_block((1, 7), c7, c7; pad=(0, 3))...)
+    branch7x7_3 = Chain(conv_block((7, 1), c7, 192; pad=(3, 0))...)
+
+    branch7x7dbl_1 = Chain(conv_block((1, 1), inplanes, c7)...)
+    branch7x7dbl_2 = Chain(conv_block((7, 1), c7, c7; pad=(3, 0))...)
+    branch7x7dbl_3 = Chain(conv_block((1, 7), c7, c7; pad=(0, 3))...)
+    branch7x7dbl_4 = Chain(conv_block((7, 1), c7, c7; pad=(3, 0))...)
+    branch7x7dbl_5 = Chain(conv_block((1, 7), c7, 192; pad=(0, 3))...)
+
+    branch_pool = Chain(MeanPool((3, 3), pad=1, stride=1), 
+                        conv_block((1, 1), inplanes, 192)...)
+
+    layer = x -> begin
+        y1 = branch1x1(x)
+
+        y2 = branch7x7_1(x)
+        y2 = branch7x7_2(y2)
+        y2 = branch7x7_3(y2)
+
+        y3 = branch7x7dbl_1(x)
+        y3 = branch7x7dbl_2(y3)
+        y3 = branch7x7dbl_3(y3)
+        y3 = branch7x7dbl_4(y3)
+        y3 = branch7x7dbl_5(y3)
+
+        y4 = branch_pool(x)
+
+        return cat(y1, y2, y3, y4; dims=3)
+    end
+end
+
+function inception_d(inplanes)
+    branch3x3_1 = Chain(conv_block((1, 1), inplanes, 192)...)
+    branch3x3_2 = Chain(conv_block((3, 3), 192, 320; stride=2)...)
+
+    branch7x7x3_1 = Chain(conv_block((1, 1), inplanes, 192)...)
+    branch7x7x3_2 = Chain(conv_block((1, 7), 192, 192; pad=(0, 3))...)
+    branch7x7x3_3 = Chain(conv_block((7, 1), 192, 192; pad=(3, 0))...)
+    branch7x7x3_4 = Chain(conv_block((3, 3), 192, 192; stride=2)...)
+
+    branch_pool = Chain(MaxPool((3, 3), stride=2))
+
+    layer = x -> begin
+        y1 = branch3x3_1(x)
+        y1 = branch3x3_2(y1)
+
+        y2 = branch7x7x3_1(x)
+        y2 = branch7x7x3_2(y2)
+        y2 = branch7x7x3_3(y2)
+        y2 = branch7x7x3_4(y2)
+
+        y3 = branch_pool(x)
+
+        return cat(y1, y2, y3; dims=3)
+    end
+end
+
+function inception_e(inplanes)
+    branch1x1 = Chain(conv_block((1, 1), inplanes, 320)...)
+
+    branch3x3_1 = Chain(conv_block((1, 1), inplanes, 384)...)
+    branch3x3_2a = Chain(conv_block((1, 3), 384, 384; pad=(0, 1))...)
+    branch3x3_2b = Chain(conv_block((3, 1), 384, 384; pad=(1, 0))...)
+
+    branch3x3dbl_1 = Chain(conv_block((1, 1), inplanes, 448)...)
+    branch3x3dbl_2 = Chain(conv_block((3, 3), 448, 384; pad=1)...)
+    branch3x3dbl_3a = Chain(conv_block((1, 3), 384, 384; pad=(0, 1))...)
+    branch3x3dbl_3b = Chain(conv_block((3, 1), 384, 384; pad=(1, 0))...)
+
+    branch_pool = Chain(MeanPool((3, 3), pad=1, stride=1),
+                        conv_block((1, 1), inplanes, 192)...)
+
+    layer = x -> begin
+        y1 = branch1x1(x)
+
+        y2 = branch3x3_1(x)
+        y2 = cat(branch3x3_2a(y2), branch3x3_2b(y2); dims=3)
+
+        y3 = branch3x3dbl_1(x)
+        y3 = branch3x3dbl_2(y3)
+        y3 = cat(branch3x3dbl_3a(y3), branch3x3dbl_3b(y3); dims=3)
+
+        y4 = branch_pool(x)
+
+        return cat(y1, y2, y3, y4; dims=3)
+    end
+end
+
+function inception3()
+    layer = Chain(conv_block((3, 3), 3, 32; stride=2)...,
+                conv_block((3, 3), 32, 32)...,
+                conv_block((3, 3), 32, 64; pad=1)...,
+                MaxPool((3, 3), stride=2),
+                conv_block((1, 1), 64, 80)...,
+                conv_block((3, 3), 80, 192)...,
+                MaxPool((3, 3), stride=2),
+                inception_a(192, 32),
+                inception_a(256, 64),
+                inception_a(288, 64),
+                inception_b(288),
+                inception_c(768, 128),
+                inception_c(768, 160),
+                inception_c(768, 160),
+                inception_c(768, 192),
+                inception_d(768),
+                inception_e(1280),
+                inception_e(2048),
+                AdaptiveMeanPool((1, 1)),
+                Dropout(0.2),
+                flatten,
+                Dense(2048, 1000))
+    Flux.testmode!(layer, false)
+    return layer
+end

test/VisionModel/inception.jl

@@ -0,0 +1,4 @@
+@testset "InceptionV3" begin
+    m = inception3()
+    @test size(m(rand(Float32, 299, 299, 3, 50))) == (1000, 50)
+end

test/runtests.jl

@@ -7,4 +7,5 @@ using Test
     include("VisionModel/resnet.jl")
     include("VisionModel/vgg.jl")
     include("VisionModel/googlenet.jl")
+    include("VisionModel/inception.jl")
 end