Sat, 20 Apr 2024 12:42:22 +0300
removed unnecessary file
################## # Our main module ################## __precompile__() module PredictPDPS ######################## # Load external modules ######################## using Printf using FileIO #using JLD2 using Setfield using ImageQualityIndexes: assess_psnr, assess_ssim using DelimitedFiles import GR using AlgTools.Util using AlgTools.StructTools using AlgTools.LinkedLists using AlgTools.Comms using ImageTools.Visualise: secs_ns, grayimg, do_visualise using ImageTools.ImFilter: gaussian ##################### # Load local modules ##################### include("OpticalFlow.jl") include("Radon.jl") include("ImGenerate.jl") include("Algorithm.jl") include("AlgorithmBoth.jl") include("AlgorithmBothGreedyV.jl") include("AlgorithmBothCumul.jl") include("AlgorithmBothMulti.jl") include("AlgorithmBothNL.jl") include("AlgorithmFB.jl") include("AlgorithmFBDual.jl") # Additional include("AlgorithmProximal.jl") include("AlgorithmGreedy.jl") include("AlgorithmRotation.jl") include("AlgorithmNoPrediction.jl") include("AlgorithmPrimalOnly.jl") include("AlgorithmDualScaling.jl") include("PET/PET.jl") import .Algorithm, .AlgorithmBoth, .AlgorithmBothGreedyV, .AlgorithmBothCumul, .AlgorithmBothMulti, .AlgorithmBothNL, .AlgorithmFB, .AlgorithmFBDual, .AlgorithmProximal, .AlgorithmGreedy, .AlgorithmRotation, .AlgorithmNoPrediction, .AlgorithmPrimalOnly, .AlgorithmDualScaling, .PET using .ImGenerate using .OpticalFlow: DisplacementFull, DisplacementConstant using .PET: demo_pet1, demo_pet2, demo_pet3, demo_pet4, demo_pet5, demo_pet6, batchrun_pet ############## # Our exports ############## export run_experiments, batchrun_article, demo_known1, demo_known2, demo_known3, demo_unknown1,demo_unknown2,demo_unknown3, batchrun_denoising, demo_denoising1, demo_denoising2, demo_denoising3, demo_denoising4, demo_denoising5, demo_denoising6, demo_pet1, demo_pet2, demo_pet3, demo_pet4, demo_pet5, demo_pet6, batchrun_pet, batchrun_predictors ################################### # Parameterisation and experiments ################################### struct Experiment mod :: Module DisplacementT :: Type imgen :: ImGen params :: NamedTuple end function Base.show(io::IO, e::Experiment) displacementname(::Type{DisplacementFull}) = "DisplacementFull" displacementname(::Type{DisplacementConstant}) = "DisplacementConstant" print(io, " mod: $(e.mod) DisplacementT: $(displacementname(e.DisplacementT)) imgen: $(e.imgen.name) $(e.imgen.dim[1])×$(e.imgen.dim[2]) params: $(e.params ⬿ (kernel = "(not shown)",)) ") end const default_save_prefix="img/" const default_params = ( ρ = 0, verbose_iter = 100, maxiter = 10000, save_results = true, save_images = true, save_images_iters = Set([1, 2, 3, 5, 10, 25, 30, 50, 100, 250, 300, 500, 1000, 2500, 3000, 5000, 10000]), pixelwise_displacement=false, dual_flow = true, prox_predict = true, handle_interrupt = true, init = :zero, plot_movement = false, ) const square = imgen_square((200, 300)) const lighthouse = imgen_shake("lighthouse", (200, 300)) const p_known₀ = ( noise_level = 0.5, shake_noise_level = 0.05, shake = 2, α = 1, ρ̃₀ = 1, σ̃₀ = 1, δ = 0.9, σ₀ = 1, τ₀ = 0.01, ) const p_unknown₀ = ( noise_level = 0.3, shake_noise_level = 0.05, shake = 2, α = 0.2, ρ̃₀ = 1, σ̃₀ = 1, σ₀ = 1, δ = 0.9, λ = 1, θ = (300*200)*100^3, kernel = gaussian((3, 3), (11, 11)), timestep = 0.5, displacement_count = 100, τ₀ = 0.01, ) const p_known₀_denoising = ( noise_level = 0.5, shake_noise_level = 0.025, shake = 3.0, α = 1.0, ρ̃₀ = 1.0, σ̃₀ = 1.0, δ = 0.9, σ₀ = 0.1, τ₀ = 0.01, ) const experiments_pdps_known = ( Experiment(Algorithm, DisplacementConstant, lighthouse, p_known₀_denoising ⬿ (phantom_ρ = 0,)), Experiment(Algorithm, DisplacementConstant, lighthouse, p_known₀ ⬿ (phantom_ρ = 100,)), Experiment(Algorithm, DisplacementConstant, square, p_known₀ ⬿ (phantom_ρ = 0,)) ) const experiments_pdps_unknown_multi = ( Experiment(AlgorithmBothMulti, DisplacementConstant, lighthouse, p_unknown₀ ⬿ (phantom_ρ = 0,)), Experiment(AlgorithmBothMulti, DisplacementConstant, lighthouse, p_unknown₀ ⬿ (phantom_ρ = 100,)), Experiment(AlgorithmBothMulti, DisplacementConstant, square, p_unknown₀ ⬿ (phantom_ρ = 0,)), ) const experiments_fb_known = ( Experiment(AlgorithmFB, DisplacementConstant, lighthouse, p_known₀ ⬿ (τ̃₀=0.9, fb_inner_iterations = 10)), ) const experiments_all = Iterators.flatten(( experiments_pdps_known, experiments_pdps_unknown_multi, experiments_fb_known )) const denoising_experiments_pdps_known = ( Experiment(AlgorithmDualScaling, DisplacementConstant, lighthouse, p_known₀_denoising), Experiment(AlgorithmGreedy, DisplacementConstant, lighthouse, p_known₀_denoising), Experiment(AlgorithmNoPrediction, DisplacementConstant, lighthouse, p_known₀_denoising), Experiment(AlgorithmPrimalOnly, DisplacementConstant, lighthouse, p_known₀_denoising), Experiment(AlgorithmProximal, DisplacementConstant, lighthouse, p_known₀_denoising ⬿ (phantom_ρ = 100,)), Experiment(AlgorithmRotation, DisplacementConstant, lighthouse, p_known₀_denoising), ) const denoising_experiments_all = Iterators.flatten(( denoising_experiments_pdps_known, )) ################ # Log ################ struct LogEntry <: IterableStruct iter :: Int time :: Float64 function_value :: Float64 #v_cumul_true_y :: Float64 #v_cumul_true_x :: Float64 #v_cumul_est_y :: Float64 #v_cumul_est_x :: Float64 psnr :: Float64 ssim :: Float64 #psnr_data :: Float64 #ssim_data :: Float64 end struct LogEntryHiFi <: IterableStruct iter :: Int v_cumul_true_y :: Float64 v_cumul_true_x :: Float64 end ############### # Main routine ############### struct State vis :: Union{Channel,Bool,Nothing} start_time :: Union{Real,Nothing} wasted_time :: Real log :: LinkedList{LogEntry} log_hifi :: LinkedList{LogEntryHiFi} aborted :: Bool end function name(e::Experiment, p) ig = e.imgen # return "$(ig.name)_$(e.mod.identifier)_$(@sprintf "%x" hash(p))" return "$(ig.name)_$(e.mod.identifier)_$(Int64(10000*p.σ₀))_$(Int64(10000*p.τ₀))" end function write_tex(texfile, e_params) open(texfile, "w") do io wp = (n, v) -> println(io, "\\def\\EXPPARAM$(n){$(v)}") wf = (n, s) -> if isdefined(e_params, s) wp(n, getfield(e_params, s)) end wf("alpha", :α) wf("sigmazero", :σ₀) wf("tauzero", :τ₀) wf("tildetauzero", :τ̃₀) wf("delta", :δ) wf("lambda", :λ) wf("theta", :θ) wf("maxiter", :maxiter) wf("noiselevel", :noise_level) wf("shakenoiselevel", :shake_noise_level) wf("shake", :shake) wf("timestep", :timestep) wf("displacementcount", :displacementcount) wf("phantomrho", :phantom_ρ) if isdefined(e_params, :σ₀) wp("sigma", (e_params.σ₀ == 1 ? "" : "$(e_params.σ₀)") * "\\sigma_{\\max}") end end end function run_experiments(;visualise=true, recalculate=true, experiments, save_prefix=default_save_prefix, fullscreen=false, kwargs...) # Create visualisation if visualise rc = Channel(1) visproc = Threads.@spawn bg_visualise_enhanced(rc, fullscreen=fullscreen) bind(rc, visproc) vis = rc else vis = false end # Run all experiments for e ∈ experiments # Parameters for this experiment e_params = default_params ⬿ e.params ⬿ kwargs ename = name(e, e_params) e_params = e_params ⬿ (save_prefix = save_prefix * ename, dynrange = e.imgen.dynrange, Λ = e.imgen.Λ) if recalculate || !isfile(e_params.save_prefix * ".txt") println("Running experiment \"$(ename)\"") # Start data generation task datachannel = Channel{OnlineData{e.DisplacementT}}(2) gentask = Threads.@spawn e.imgen.f(e.DisplacementT, datachannel, e_params) bind(datachannel, gentask) # Run algorithm iterate = curry(iterate_visualise, datachannel, State(vis, nothing, 0.0, nothing, nothing, false)) x, y, st = e.mod.solve(e.DisplacementT; dim=e.imgen.dim, iterate=iterate, params=e_params) # Clear non-saveable things st = @set st.vis = nothing println("Wasted_time: $(st.wasted_time)s") if e_params.save_results println("Saving " * e_params.save_prefix * "(.txt,_hifi.txt,_params.tex)") perffile = e_params.save_prefix * ".txt" hififile = e_params.save_prefix * "_hifi.txt" texfile = e_params.save_prefix * "_params.tex" # datafile = e_params.save_prefix * ".jld2" write_log(perffile, st.log, "# params = $(e_params)\n") write_log(hififile, st.log_hifi, "# params = $(e_params)\n") write_tex(texfile, e_params) # @save datafile x y st params end close(datachannel) wait(gentask) if st.aborted break end else println("Skipping already computed experiment \"$(ename)\"") # texfile = e_params.save_prefix * "_params.tex" # write_tex(texfile, e_params) end end if visualise # Tell subprocess to finish, and wait put!(rc, nothing) close(rc) wait(visproc) end return end ####################### # Demos and batch runs ####################### function demo(experiment; kwargs...) run_experiments(;experiments=(experiment,), save_results=false, save_images=false, visualise=true, recalculate=true, verbose_iter=1, fullscreen=true, kwargs...) end demo_known1 = () -> demo(experiments_pdps_known[3]) demo_known2 = () -> demo(experiments_pdps_known[1]) demo_known3 = () -> demo(experiments_pdps_known[2]) demo_unknown1 = () -> demo(experiments_pdps_unknown_multi[3], plot_movement=true) demo_unknown2 = () -> demo(experiments_pdps_unknown_multi[1], plot_movement=true) demo_unknown3 = () -> demo(experiments_pdps_unknown_multi[2], plot_movement=true) demo_denoising1 = () -> demo(denoising_experiments_pdps_known[1]) # Dual scaling demo_denoising2 = () -> demo(denoising_experiments_pdps_known[2]) # Greedy demo_denoising3 = () -> demo(denoising_experiments_pdps_known[3]) # No Prediction demo_denoising4 = () -> demo(denoising_experiments_pdps_known[4]) # Primal Only demo_denoising5 = () -> demo(denoising_experiments_pdps_known[5]) # Proximal (old) demo_denoising6 = () -> demo(denoising_experiments_pdps_known[6]) # Rotation function batchrun_article(kwargs...) run_experiments(;experiments=experiments_all, save_results=true, save_images=true, visualise=false, recalculate=false, kwargs...) end function batchrun_denoising(kwargs...) run_experiments(;experiments=denoising_experiments_all, save_results=true, save_images=true, visualise=false, recalculate=false, kwargs...) end function batchrun_predictors(kwargs...) batchrun_denoising() batchrun_pet() end ###################################################### # Iterator that does visualisation and log collection ###################################################### function iterate_visualise(datachannel::Channel{OnlineData{DisplacementT}}, st :: State, step :: Function, params :: NamedTuple) where DisplacementT try sc = nothing d = take!(datachannel) for iter=1:params.maxiter dnext = take!(datachannel) st = step(d.b_noisy, d.v, dnext.b_noisy) do calc_objective stn = st if isnothing(stn.start_time) # The Julia precompiler is a miserable joke, apparently not crossing module # boundaries, so only start timing after the first iteration. stn = @set stn.start_time=secs_ns() end verb = params.verbose_iter!=0 && mod(iter, params.verbose_iter) == 0 # Normalise movement to image dimensions so # our TikZ plotting code doesn't need to know # the image pixel size. sc = 1.0./maximum(size(d.b_true)) if verb || iter ≤ 20 || (iter ≤ 200 && mod(iter, 10) == 0) verb_start = secs_ns() tm = verb_start - stn.start_time - stn.wasted_time value, x, v, vhist = calc_objective() entry = LogEntry(iter, tm, value, #sc*d.v_cumul_true[1], #sc*d.v_cumul_true[2], #sc*v[1], sc*v[2], assess_psnr(x, d.b_true), assess_ssim(x, d.b_true), #assess_psnr(d.b_noisy, d.b_true), #assess_ssim(d.b_noisy, d.b_true) ) # (**) Collect a singly-linked list of log to avoid array resizing # while iterating stn = @set stn.log=LinkedListEntry(entry, stn.log) if !isnothing(vhist) vhist=vhist.*sc end if verb @printf("%d/%d J=%f, PSNR=%f, SSIM=%f, avg. FPS=%f\n", iter, params.maxiter, value, entry.psnr, entry.ssim, entry.iter/entry.time) if isa(stn.vis, Channel) put_onlylatest!(stn.vis, ((d.b_noisy, x), params.plot_movement, stn.log, vhist)) end end if params.save_images && (!haskey(params, :save_images_iters) || iter ∈ params.save_images_iters) fn = (t, ext) -> "$(params.save_prefix)_$(t)_frame$(iter).$(ext)" # save(File(format"PNG", fn("true", "png")), grayimg(d.b_true)) # save(File(format"PNG", fn("data", "png")), grayimg(d.b_noisy)) save(File(format"PNG", fn("reco", "png")), grayimg(x)) if !isnothing(vhist) open(fn("movement", "txt"), "w") do io writedlm(io, ["est_y" "est_x"]) writedlm(io, vhist) end end end stn = @set stn.wasted_time += (secs_ns() - verb_start) return stn end hifientry = LogEntryHiFi(iter, sc*d.v_cumul_true[1], sc*d.v_cumul_true[2]) st = @set st.log_hifi=LinkedListEntry(hifientry, st.log_hifi) return st end d=dnext end catch ex if params.handle_interrupt && isa(ex, InterruptException) # If SIGINT is received (user pressed ^C), terminate computations, # returning current status. Effectively, we do not call `step()` again, # ending the iterations, but letting the algorithm finish up. # Assuming (**) above occurs atomically, `st.log` should be valid, but # any results returned by the algorithm itself may be partial, as for # reasons of efficiency we do *not* store results of an iteration until # the next iteration is finished. printstyled("\rUser interrupt—finishing up.\n", bold=true, color=202) st = @set st.aborted = true else rethrow(ex) end end return st end function bg_visualise_enhanced(rc; fullscreen=false) process_channel(rc) do d imgs, plot_movement, log, vhist = d do_visualise(imgs, refresh=false, fullscreen=fullscreen) # Overlay movement GR.settextcolorind(5) GR.setcharheight(0.015) GR.settextpath(GR.TEXT_PATH_RIGHT) tx, ty = GR.wctondc(0, 1) GR.text(tx, ty, @sprintf "FPS %.1f, SSIM %.2f, PSNR %.1f" (log.value.iter/log.value.time) log.value.ssim log.value.psnr) if plot_movement sc=1.0 p=unfold_linked_list(log) x=map(e -> 1.5+sc*e.v_cumul_true_x, p) y=map(e -> 0.5+sc*e.v_cumul_true_y, p) GR.setlinewidth(2) GR.setlinecolorind(2) GR.polyline(x, y) x=map(e -> 1.5+sc*e.v_cumul_est_x, p) y=map(e -> 0.5+sc*e.v_cumul_est_y, p) GR.setlinecolorind(3) GR.polyline(x, y) if vhist != nothing GR.setlinecolorind(4) x=map(v -> 1.5+sc*v, vhist[:,2]) y=map(v -> 0.5+sc*v, vhist[:,1]) GR.polyline(x, y) end end GR.updatews() end end ############### # Precompiling ############### # precompile(Tuple{typeof(GR.drawimage), Float64, Float64, Float64, Float64, Int64, Int64, Array{UInt32, 2}}) # precompile(Tuple{Type{Plots.Plot{T} where T<:RecipesBase.AbstractBackend}, Plots.GRBackend, Int64, Base.Dict{Symbol, Any}, Base.Dict{Symbol, Any}, Array{Plots.Series, 1}, Nothing, Array{Plots.Subplot{T} where T<:RecipesBase.AbstractBackend, 1}, Base.Dict{Any, Plots.Subplot{T} where T<:RecipesBase.AbstractBackend}, Plots.EmptyLayout, Array{Plots.Subplot{T} where T<:RecipesBase.AbstractBackend, 1}, Bool}) # precompile(Tuple{typeof(Plots._plot!), Plots.Plot{Plots.GRBackend}, Base.Dict{Symbol, Any}, Tuple{Array{ColorTypes.Gray{Float64}, 2}}}) end # Module