Mercurial > repos > PredictPDPS / file comparison
comparison: src/OpticalFlow.jl
src/OpticalFlow.jl
- changeset 50
- b413b7df8cd6
- parent 48
- 06f95cdd9abe
- child 52
- cb029cdb141a
equal
deleted
inserted
replaced
| 48:06f95cdd9abe | 50:b413b7df8cd6 |
|---|---|
| 196 oldx = copy(x) | 196 oldx = copy(x) |
| 197 flow!(x, u; threads=(threads==:inner)) | 197 flow!(x, u; threads=(threads==:inner)) |
| 198 C = similar(y) | 198 C = similar(y) |
| 199 cc = abs.(x-oldx) | 199 cc = abs.(x-oldx) |
| 200 cm = max(flow.threshold,maximum(cc)) | 200 cm = max(flow.threshold,maximum(cc)) |
| 201 c = 1 .* (1 .- cc./ cm) .^flow.exponent | 201 # c = 1 .* (1 .- cc./ cm) .^flow.exponent |
| 202 c = 1.0 .- 0.75.*activation.(cc./cm) | |
| 202 C[1,:,:] .= c | 203 C[1,:,:] .= c |
| 203 C[2,:,:] .= c | 204 C[2,:,:] .= c |
| 204 y .= C.*y | 205 y .= C.*y |
| 205 end | 206 end |
| 206 | 207 |
| 207 function activation(x) | 208 function activation(x :: Real) |
| 208 return (1/(1 + exp(-10000(x - 0.05)))) # best for shepp logan | 209 return (1/(1 + exp(-1000(x - 0.05)))) # best for shepp logan |
| 209 # return -abs(x-1)^1/5 + 1 # best for lighthouse and brainphantom | 210 #return -abs(x-1)^1/5 + 1 # best for lighthouse and brainphantom |
| 210 # return x^(5) | 211 # return x^(5) |
| 211 # return (1/(1 + exp(-1000(x - 0.075)))) | 212 # return (1/(1 + exp(-1000(x - 0.075)))) |
| 212 # return 4*(x-0.5)^3 + 0.5 | 213 # return 4*(x-0.5)^3 + 0.5 |
| 213 # return (1/(1 + exp(-1000(x - 0.05)))) | 214 # return (1/(1 + exp(-1000(x - 0.05)))) |
| 214 # return -3(x-0.5)^2 + 0.75 | 215 # return -3(x-0.5)^2 + 0.75 |
| 215 # return (x-1)^51 + 1 | 216 # return (x-1)^51 + 1 |
| 216 # return -abs(x-1)^1/3 + 1 | 217 # return -abs(x-1)^1/3 + 1 |
| 217 # return 16*(x-0.5)^5 + 0.5 | 218 # return 16*(x-0.5)^5 + 0.5 |
| 218 end | 219 end |
| 219 | 220 |
| 220 function activationD(x) | 221 |
| 221 # return (1/(1 + exp(-500(x - 0.1)))) | 222 # # Experimental predictor for dual scaling based on activation |
| 222 # return 4*(x-0.5)^3 + 0.5 | 223 # function pdflow!(x, Δx, y, Δy, u, flow :: ActivatedDual; threads=:none) |
| 223 #return (1/(1 + exp(-1000(x - 0.1)))) | 224 # @assert(size(u)==(2,)) |
| 224 return 1.0 | 225 # oldx = copy(x) |
| 225 end | 226 # flow!(x, u; threads=(threads==:inner)) |
| 226 | 227 # C = similar(y) |
| 227 # Experimental predictor for dual scaling based on activation | 228 # cc = abs.(x-oldx) |
| 228 function pdflow!(x, Δx, y, Δy, u, flow :: ActivatedDual; threads=:none) | 229 # cm = max(1e-12,maximum(cc)) |
| 229 @assert(size(u)==(2,)) | 230 # c = 1.0 .- 0.75.*activation.(cc./cm) |
| 230 oldx = copy(x) | 231 # C[1,:,:] .= c |
| 231 flow!(x, u; threads=(threads==:inner)) | 232 # C[2,:,:] .= c |
| 232 C = similar(y) | 233 # y .= C.*y |
| 233 cc = abs.(x-oldx) | 234 # #y .= y .- C.*y # 0.75 for brain phantom and shepp logan |
| 234 cm = max(1e-12,maximum(cc)) | 235 # end |
| 235 c = activation.(cc ./ cm) | |
| 236 C[1,:,:] .= c | |
| 237 C[2,:,:] .= c | |
| 238 # y .= C.*y | |
| 239 y .= y .- 0.75.*C.*y # 0.75 for brain phantom and shepp logan | |
| 240 end | |
| 241 | 236 |
| 242 function pdflow!(x, Δx, y, Δy, u, :: ZeroDual; threads=:none) | 237 function pdflow!(x, Δx, y, Δy, u, :: ZeroDual; threads=:none) |
| 243 @assert(size(u)==(2,)) | 238 @assert(size(u)==(2,)) |
| 244 flow!(x, u; threads=(threads==:inner)) | 239 flow!(x, u; threads=(threads==:inner)) |
| 245 y .= 0.0 | 240 y .= 0.0 |
| 304 Δx = warp(x, tform, axes(x), fillvalue=Flat()) | 299 Δx = warp(x, tform, axes(x), fillvalue=Flat()) |
| 305 @. x = Δx | 300 @. x = Δx |
| 306 C = similar(y) | 301 C = similar(y) |
| 307 cc = abs.(x-oldx) | 302 cc = abs.(x-oldx) |
| 308 cm = max(flow.threshold,maximum(cc)) | 303 cm = max(flow.threshold,maximum(cc)) |
| 309 c = 1 .* (1 .- cc./ cm) .^flow.exponent | 304 # c = 1 .* (1 .- cc./ cm) .^flow.exponent # Original dual scaling |
| 305 # c = 1.0 .- 0.75.*activation.(cc./cm) # Best for brain phantom | |
| 306 c = 1.0 .- 1.0.*activation.(cc./cm) # Best for shepp logan phantom | |
| 310 C[1,:,:] .= c | 307 C[1,:,:] .= c |
| 311 C[2,:,:] .= c | 308 C[2,:,:] .= c |
| 312 y .= C.*y | 309 y .= C.*y |
| 313 end | 310 end |
| 314 | 311 |
| 315 # Experimental predictor for dual scaling based on activation | 312 # # Experimental predictor for dual scaling based on activation |
| 316 function petpdflow!(x, Δx, y, Δy, u, theta_known, flow :: ActivatedDual; threads=:none) | 313 # function petpdflow!(x, Δx, y, Δy, u, theta_known, flow :: ActivatedDual; threads=:none) |
| 317 oldx = copy(x) | 314 # oldx = copy(x) |
| 318 center_point = center(x) .+ u | 315 # center_point = center(x) .+ u |
| 319 tform = recenter(RotMatrix(theta_known[1]), center_point) | 316 # tform = recenter(RotMatrix(theta_known[1]), center_point) |
| 320 Δx = warp(x, tform, axes(x), fillvalue=Flat()) | 317 # Δx = warp(x, tform, axes(x), fillvalue=Flat()) |
| 321 @. x = Δx | 318 # @. x = Δx |
| 322 C = similar(y) | 319 # C = similar(y) |
| 323 cc = abs.(x-oldx) | 320 # cc = abs.(x-oldx) |
| 324 cm = max(1e-12,maximum(cc)) | 321 # cm = max(1e-12,maximum(cc)) |
| 325 c = activation.(cc ./ cm) | 322 # c = activation.(cc ./ cm) |
| 326 C[1,:,:] .= c | 323 # C[1,:,:] .= c |
| 327 C[2,:,:] .= c | 324 # C[2,:,:] .= c |
| 328 # Δx .= activation.(sqrt.(abs.(y[1,:,:]).^2 + abs.(y[2,:,:]).^2))./0.25 | 325 # # Δx .= activation.(sqrt.(abs.(y[1,:,:]).^2 + abs.(y[2,:,:]).^2))./0.25 |
| 329 # D = similar(y) | 326 # # D = similar(y) |
| 330 # D[1,:,:] .= Δx | 327 # # D[1,:,:] .= Δx |
| 331 # D[2,:,:] .= Δx | 328 # # D[2,:,:] .= Δx |
| 332 # y .= C.*y | 329 # # y .= C.*y |
| 333 # y .= y .- 1.0.*C.*D.*y | 330 # # y .= y .- 1.0.*C.*D.*y |
| 334 y .= y .- 1.0.*C.*y | 331 # #y .= y .- 1.0.*C.*y |
| 335 end | 332 # end |
| 336 | 333 |
| 337 # Method for rotation prediction (exploiting property of inverse rotation) | 334 # Method for rotation prediction (exploiting property of inverse rotation) |
| 338 function petpdflow!(x, Δx, y, Δy, u, theta_known, flow :: Rotation; threads=:none) | 335 function petpdflow!(x, Δx, y, Δy, u, theta_known, flow :: Rotation; threads=:none) |
| 339 @backgroundif (threads==:outer) begin | 336 @backgroundif (threads==:outer) begin |
| 340 petflow!(x, Δx, u, theta_known; threads=(threads==:inner)) | 337 petflow!(x, Δx, u, theta_known; threads=(threads==:inner)) |
