Training set

15 image pairs with ground truth, shown below at 10% of their original size. Trailing letters indicate dataset types:
P - perfect rectification (imperfect by default);
E - changed exposure between views;
L - changed lighting between views.

Mouse over the input images to flip between views.

Dataset	View 0/1	GT disps 0/1	GT y-disps
Adirondack 5.7 Mpix ndisp=290 dymax=1.7
ArtL 1.5 Mpix ndisp=256
Jadeplant 5.2 Mpix ndisp=640 dymax=1.4
Motorcycle 5.9 Mpix ndisp=280 dymax=1.5
MotorcycleE 5.9 Mpix ndisp=280 dymax=1.5
Piano 5.4 Mpix ndisp=260 dymax=1.9
PianoL 5.4 Mpix ndisp=260 dymax=1.9
Pipes 5.7 Mpix ndisp=300 dymax=5.2
Playroom 5.3 Mpix ndisp=330 dymax=0.9
Playtable 5.0 Mpix ndisp=290 dymax=2.7
PlaytableP 5.0 Mpix ndisp=290
Recycle 5.6 Mpix ndisp=260 dymax=1.5
Shelves 5.9 Mpix ndisp=240 dymax=1.3
Teddy 2.7 Mpix ndisp=256
Vintage 5.5 Mpix ndisp=760 dymax=1.4