GaussianBlur and ColorJitter ops.

Project.toml

@@ -18,6 +18,7 @@ MLDataPattern = "9920b226-0b2a-5f5f-9153-9aa70a013f8b"
 OffsetArrays = "6fe1bfb0-de20-5000-8ca7-80f57d26f881"
 Rotations = "6038ab10-8711-5258-84ad-4b1120ba62dc"
 StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
+Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 
 [compat]
 ComputationalResources = "0.3"

docs/operations/blur/config.json

@@ -0,0 +1,6 @@
+{
+  "title": "Blurring",
+  "order": [
+    "gaussianblur.jl"
+  ]
+}

docs/operations/blur/gaussianblur.jl

@@ -0,0 +1,34 @@
+# ---
+# title: GaussianBlur
+# cover: gaussianblur.gif
+# description: blur the input image using a gaussian kernel
+# ---
+
+# [`GaussianBlur`](@ref) can be used to blur the input image using a gaussian
+# kernel with a specified kernel size and standard deviation.
+
+using Augmentor
+using ImageShow, ImageCore
+
+img_in = testpattern(RGB, ratio=0.5)
+
+mosaicview(
+    img_in,
+    augment(img_in, GaussianBlur(3)),
+    augment(img_in, GaussianBlur(5, 2.5));
+    fillvalue=colorant"white", nrow=1, npad=10
+)
+
+# ## References
+
+#md # ```@docs
+#md # GaussianBlur
+#md # ```
+
+
+## save covers #src
+using ImageMagick #src
+using FileIO #src
+include(joinpath("..", "assets", "utilities.jl")) #src
+cover = make_gif(testpattern(RGB, ratio=0.5), GaussianBlur(5), 2) #src
+ImageMagick.save("gaussianblur.gif", cover; fps=1) #src

docs/operations/color/colorjitter.jl

@@ -0,0 +1,32 @@
+# ---
+# title: ColorJitter
+# cover: colorjitter.gif
+# description: Adjust contrast and brightness of an image
+# ---
+
+# [`ColorJitter`](@ref) can be used to adjust the contrast and brightness of an input image.
+
+using Augmentor
+using ImageShow, ImageCore
+
+img_in = testpattern(RGB, ratio=0.5)
+
+mosaicview(
+    img_in,
+    augment(img_in, ColorJitter(1.2, 0.3)),
+    augment(img_in, ColorJitter(0.75, -0.2));
+    fillvalue=colorant"white", nrow=1, npad=10
+)
+
+# ## References
+
+#md # ```@docs
+#md # ColorJitter
+#md # ```
+
+## save covers #src
+using ImageMagick #src
+using FileIO #src
+include(joinpath("..", "assets", "utilities.jl")) #src
+cover = make_gif(testpattern(RGB, ratio=0.5), ColorJitter(), 4) #src
+ImageMagick.save("colorjitter.gif", cover; fps=1) #src

docs/operations/color/config.json

@@ -0,0 +1,7 @@
+{
+  "title": "Color adjustments",
+  "order": [
+    "colorjitter.jl"
+  ],
+  "description": ""
+}

docs/operations/config.json

@@ -2,8 +2,10 @@
   "order": [
     "affine",
     "distortions",
+    "color",
+    "blur",
     "size",
     "misc"
   ]
 }
-
+

src/Augmentor.jl

@@ -69,6 +69,9 @@ export
 
     ElasticDistortion,
 
+    ColorJitter,
+    GaussianBlur,
+
     augment,
     augment!,
     augmentbatch!,
@@ -95,6 +98,8 @@ include("operations/resize.jl")
 include("operations/scale.jl")
 include("operations/zoom.jl")
 include("operations/either.jl")
+include("operations/color.jl")
+include("operations/blur.jl")
 
 include("distortionfields.jl")
 include("distortedview.jl")

src/operations/blur.jl

@@ -0,0 +1,75 @@
+import ImageFiltering: imfilter, KernelFactors.gaussian
+
+"""
+    GaussianBlur <: ImageOperation
+
+Description
+--------------
+
+Blurs an image using a Gaussian filter.
+
+Usage
+--------------
+
+    GaussianBlur(k, [σ])
+
+Arguments
+--------------
+
+- **`k`** : `Integer` or `AbstractVector` of `Integer` that denote
+    the kernel size. It must be an odd positive number.
+- **`σ`** : Optional. `Real` or `AbstractVector` of `Real` that denote the
+    standard deviation. It must be a positive number.
+    Defaults to `0.3 * ((k - 1) / 2 - 1) + 0.8`.
+
+Examples
+--------------
+
+```
+using Augmentor
+img = testpattern()
+
+# use exactly k=3 and σ=1.0
+augment(img, GaussianBlur(3, 1.0))
+
+# pick k and σ randomly from the specified ranges
+augment(img, GaussianBlur(3:2:7, 1.0:0.1:2.0))
+```
+"""
+struct GaussianBlur{K <: AbstractVector, S <: AbstractVector} <: ImageOperation
+    k::K
+    σ::S
+
+    function GaussianBlur(k::K, σ::S) where {K <: AbstractVector{<:Integer},
+                                             S <: AbstractVector{<:Real}}
+        minimum(k) > 0 || throw(ArgumentError("Kernel size must be positive: $(k)"))
+        minimum(σ) > 0 || throw(ArgumentError("σ must be positive: $(σ)"))
+        new{K, S}(k, σ)
+    end
+end
+
+GaussianBlur(k, σ) = GaussianBlur(vectorize(k), vectorize(σ))
+GaussianBlur(k) = GaussianBlur(k, 0.3 * ((k - 1) / 2 - 1) + 0.8)
+
+randparam(op::GaussianBlur, img) = (safe_rand(op.k), safe_rand(op.σ))
+
+@inline supports_eager(::Type{<:GaussianBlur}) = true
+
+function applyeager(op::GaussianBlur, img::AbstractArray, (k, σ))
+    kernel = gaussian((σ, σ), (k, k))
+    return imfilter(img, kernel)
+end
+
+function showconstruction(io::IO, op::GaussianBlur)
+    print(io, typeof(op).name.name, '(', op.k, ", ", op.σ,')')
+end
+
+function Base.show(io::IO, op::GaussianBlur)
+    if get(io, :compact, false)
+        print(io, "GaussianBlur with k=$(op.k) and σ=$(op.σ)")
+    else
+        print(io, "Augmentor.")
+        showconstruction(io, op)
+    end
+end
+

src/operations/color.jl

@@ -0,0 +1,117 @@
+import ImageCore: clamp01, gamutmax
+import Statistics: mean
+
+"""
+    ColorJitter <: ImageOperation
+
+Description
+--------------
+
+Adjusts the brightness and contrast of an image according to the formula
+`α * image[i] + β * M`, where `M` is either `mean(image)` or the maximum
+intensity value.
+
+Usage
+--------------
+
+    ColorJitter()
+    ColorJitter(α, β; [usemax])
+
+Arguments
+--------------
+
+- **`α`** : `Real` or `AbstractVector` of `Real` that denote the coefficient(s)
+    for contrast adjustment. Defaults to `0.8:0.1:1.2`.
+- **`β`** : `Real` or `AbstractVector` of `Real` that denote the coefficient(s)
+    for brightness adjustment. Defaults to `-0.2:0.1:0.2`.
+- **`usemax::Bool`**: Optional. If `true`, the brightness will be adjusted by
+    the maximum intensity value; otherwise, the image mean will be used.
+    Defaults to `true`.
+
+Examples
+--------------
+
+```
+using Augmentor
+img = testpattern()
+
+# use exactly 1.2 for contrast, and one of 0.5 and 0.8 for brightness
+augment(img, ColorJitter(1.2, [0.5, 0.8]))
+
+# pick the coefficients randomly from the specified ranges
+augment(img, ColorJitter(0.8:0.1:2.0, 0.5:0.1:1.1))
+```
+"""
+struct ColorJitter{A<:AbstractVector, B<:AbstractVector} <: ImageOperation
+    α::A
+    β::B
+    usemax::Bool
+
+    function ColorJitter(α::A, β::B, usemax) where {A<:AbstractVector{<:Real},
+                                                    B<:AbstractVector{<:Real}}
+        length(α) > 0 || throw(ArgumentError("Range $(α) is empty"))
+        length(β) > 0 || throw(ArgumentError("Range $(β) is empty"))
+        new{A, B}(α, β, usemax)
+    end
+end
+
+ColorJitter() = ColorJitter(0.8:0.1:1.2, -0.2:0.1:0.2, true)
+ColorJitter(α, β; usemax=true) = ColorJitter(vectorize(α), vectorize(β), usemax)
+
+randparam(op::ColorJitter, img) = (safe_rand(op.α), safe_rand(op.β))
+
+@inline supports_eager(::Type{<:ColorJitter}) = true
+@inline supports_lazy(::Type{<:ColorJitter}) = true
+
+function applyeager(op::ColorJitter, img::AbstractArray, (α, β))
+    M = _get_M(op, img)
+    return _map_pix.(α, β, M, img)
+end
+
+function applylazy(op::ColorJitter, img::AbstractArray, (α, β))
+    M = _get_M(op, img)
+    return ColorJitterView(img, α, β, M)
+end
+
+function showconstruction(io::IO, op::ColorJitter)
+    print(io, typeof(op).name.name, '(', op.α, ", ", op.β, ", ", op.usemax, ')')
+end
+
+function Base.show(io::IO, op::ColorJitter)
+    if get(io, :compact, false)
+        maxmsg = op.usemax ? "max. intensity" : "mean value";
+        print(io, "Color jitter with coffecients α=$(op.α) and β=$(op.β) (w.r.t. $(maxmsg))")
+    else
+        print(io, "Augmentor.")
+        showconstruction(io, op)
+    end
+end
+
+_map_pix(α, β, M, pix) = clamp01(α * pix + β * M)
+function _get_M(op::ColorJitter, img)
+    if op.usemax
+        T = eltype(img)
+        return T(gamutmax(eltype(img))...)
+    else
+        return mean(img)
+    end
+end
+
+# This wraps an image so that its pixels appear as after the contrast and
+# brightness adjustment. This is done in the `getindex` method.
+struct ColorJitterView{T, P<:AbstractMatrix{T}, R} <: AbstractArray{T, 2}
+    orig::P
+    α::R
+    β::R
+    M::T
+
+    function ColorJitterView(img::P, α::R, β::R, M) where {P <: AbstractMatrix,
+                                                        R <: Real}
+        new{eltype(P), P, R}(img, α, β, M)
+    end
+end
+
+Base.parent(A::ColorJitterView) = A.parent
+Base.size(A::ColorJitterView) = size(A.orig)
+Base.axes(A::ColorJitterView) = axes(A.orig)
+Base.getindex(A::ColorJitterView, i, j) = _map_pix(A.α, A.β, A.M, A.orig[i, j])

test/operations/tst_blur.jl

@@ -0,0 +1,34 @@
+@testset "GaussianBlur" begin
+    @testset "constructor" begin
+        @test_throws ArgumentError GaussianBlur(0)
+        @test_throws ArgumentError GaussianBlur(3, 0)
+        @test_throws ArgumentError GaussianBlur([1, 3], 0)
+        @test_throws ArgumentError GaussianBlur([1, 0], 1)
+        @test_throws ArgumentError GaussianBlur(3, -1:1)
+    end
+    @testset "randparam" begin
+        img = testpattern()
+        ks = [3, 1:2:5, [1, 3, 5]]
+        σs = [1, 1:0.1:2, [1, 2]]
+        for k in ks, σ in σs
+            op = GaussianBlur(k, σ)
+            p = @inferred Augmentor.randparam(op, img)
+            @test p[1] in k
+            @test p[2] in σ
+        end
+    end
+    @testset "eager" begin
+        @test Augmentor.supports_eager(GaussianBlur)
+
+        k = 3
+        σ = 3.2
+
+        imgs = [testpattern(), camera]
+
+        for img in imgs
+            ref = imfilter(img, KernelFactors.gaussian((σ, σ), (k, k)))
+            res = Augmentor.applyeager(GaussianBlur(k, σ), img)
+            @test ref == res
+        end
+    end
+end