Table 2: Evaluation results based on the online Middlebury stereo benchmark [4]: the errors are given in percentages for the nonoccluded (NONOCC) region, the whole image (ALL), and discontinuity (DISC) areas. The numbers within brackets indicate the ranking in the Middlebury table.

Images Nonocc All Disc

Tsukuba 1.33 (37) 1.82 (32) 7.19 (46)
Venus 0.32 (39) 0.79 (46) 4.5 (58)
Teddy 5.32 (17) 11.9 (40) 14.5 (19)
Cones 2.73 ( 𝟏 𝟒 ) 9.69 (53) 7.91 (21)