problemname TsukubaStereo1
width 384
height 288
labels 16
labelscale 16
smoothexp 1
smoothmax 2
lambda 20
datacost tsukuba/datacost%03d%c.png
datascale 65
smoothx tsukuba/smoothx%c.png
smoothy tsukuba/smoothy%c.png
smoothscale 25600

datacost000H.png  datacost004L.png  datacost009H.png  datacost013L.png
datacost000L.png  datacost005H.png  datacost009L.png  datacost014H.png
datacost001H.png  datacost005L.png  datacost010H.png  datacost014L.png
datacost001L.png  datacost006H.png  datacost010L.png  datacost015H.png
datacost002H.png  datacost006L.png  datacost011H.png  datacost015L.png
datacost002L.png  datacost007H.png  datacost011L.png  smoothXH.png
datacost003H.png  datacost007L.png  datacost012H.png  smoothXL.png
datacost003L.png  datacost008H.png  datacost012L.png  smoothYH.png
datacost004H.png  datacost008L.png  datacost013H.png  smoothYL.png

The parameters used for cost and smoothness images:

match_fn 1   # SAD, i.e., cost = |diff|
match_interval 1 # Birchfield / Tomasi

opt_grad_thresh 8 # i.e. if gradient <= 8, then
opt_grad_penalty 2 # multiply smoothness cost by 2