Table 7 The comparison of the CR-RNN architectures with respect to original methods in terms of total parameters and different error rates on KITTI 2012 test set
From: Refinement of matching costs for stereo disparities using recurrent neural networks
Architectures | B (%) | EPE | GD(%) | |||
|---|---|---|---|---|---|---|
|  | > 5px | > 4px | > 3px | > 2px |  |  |
MC-CNN-Fast | 16.03 | 16.80 | 17.91 | 20.14 | 7.4006 | 35.12 |
MC-CNN-Acrt | 13.18 | 13.81 | 14.76 | 16.86 | 6.5936 | 35.13 |
MC-CNN-Fast + CR-RNN C | 13.53 | 14.36 | 15.65 | 18.66 | 4.8488 | 26.84 |
MC-CNN-Fast + CR-RNN P | 12.89 | 13.78 | 15.13 | 18.26 | 4.4290 | 30.10 |
SAD | 31.90 | 33.44 | 35.34 | 37.59 | 13.1638 | 74.15 |
SAD + CR-RNN P | 30.51 | 31.73 | 33.40 | 35.63 | 12.9390 | 72.47 |
CBCA | 29.94 | 31.42 | 33.24 | 35.46 | 12.3728 | 68.13 |
CBCA + CR-RNN P | 25.42 | 26.75 | 28.57 | 31.10 | 10.4674 | 50.10 |