Fri, 03 May 2024 13:33:23 -0500
Change title
#################################################################### # Predictive online PDPS for optical flow with known velocity field #################################################################### __precompile__() module AlgorithmFB identifier = "fb_known" using Printf using AlgTools.Util import AlgTools.Iterate using ImageTools.Gradient using ..OpticalFlow: Image, ImageSize, flow! ######################### # Iterate initialisation ######################### function init_rest(x::Image) imdim=size(x) y = zeros(2, imdim...) Δx = copy(x) Δy = copy(y) ỹ = copy(y) y⁻ = copy(y) return x, y, Δx, Δy, ỹ, y⁻ end function init_iterates(xinit::Image) return init_rest(copy(xinit)) end function init_iterates(dim::ImageSize) return init_rest(zeros(dim...)) end ############ # Algorithm ############ function solve( :: Type{DisplacementT}; dim :: ImageSize, iterate = AlgTools.simple_iterate, params::NamedTuple) where DisplacementT ################################ # Extract and set up parameters ################################ α, ρ = params.α, params.ρ τ₀, τ̃₀ = params.τ₀, params.τ̃₀ R_K² = ∇₂_norm₂₂_est² τ̃ = τ̃₀/R_K² τ = τ₀ ###################### # Initialise iterates ###################### x, y, Δx, Δy, ỹ, y⁻ = init_iterates(dim) init_data = (params.init == :data) #################### # Run the algorithm #################### v = iterate(params) do verbose :: Function, b :: Image, v_known :: DisplacementT, 🚫unused_b_next :: Image ################## # Prediction step ################## if init_data x .= b init_data = false else # Δx is a temporary storage variable of correct dimensions flow!(x, v_known, Δx) end ################################################################## # We need to do forward–backward step on min_x |x-b|^2/2 + α|∇x|. # The forward step is easy, the prox requires solving the predual # problem of a problem similar to the original. ################################################################## @. x = x-τ*(x-b) ############## # Inner FISTA ############## t = 0 # Move step length from proximal quadratic term into L1 term. α̃ = α*τ @. ỹ = y for i=1:params.fb_inner_iterations ∇₂ᵀ!(Δx, ỹ) @. Δx .-= x ∇₂!(Δy, Δx) @. y⁻ = y @. y = (ỹ - τ̃*Δy)/(1 + τ̃*ρ/α̃) proj_norm₂₁ball!(y, α̃) t⁺ = (1+√(1+4*t^2))/2 @. ỹ = y+((t-1)/t⁺)*(y-y⁻) t = t⁺ end ∇₂ᵀ!(Δx, y) @. x = x - Δx ################################ # Give function value if needed ################################ v = verbose() do ∇₂!(Δy, x) value = norm₂²(b-x)/2 + α*γnorm₂₁(Δy, ρ) value, x, [NaN, NaN], nothing end v end return x, y, v end end # Module