Datasets
The popular datasets for stereo matching include:
- synthetic datasets: e.g., Scene Flow dataset;
- Real world datasets: e.g, Middlebury V3, KITTI 2012&2015, ETH3D.
Statistics Info
- Scene Flow: $\approx$ 39k stereo frames in $960 \times 540$ pixel resolution.
- Total Training Image # : 35454, including 4400 # Driving, 8664 # Monkaa, 22390 # Flyingthings3D;
- Training Image # with disparity $\geq 256$ : 6555, i.e., $ 6555 / 35454 = 18.49\%$
- Training Image # with disparity $\geq 192$ : 7996, i.e., $ 7996 / 35454 = 22.55\%$
- So we should exclude the pixel whose disparity is larger than 256, in the loss function for network training.
- KITTI : KITTI 12 and KITTI 15, roughtly with size $1240 \times 388$.
- ETH3D : See the following tables. According to its website, no information about the disparity range is provided: the
corresponding field is set to the image width.
- Middlebury V3: see the following tables.
Middlebury V3 Info :
| Images (including 15 images) |
Height |
Width |
Max Disparity |
vmin |
vmax |
| Adirondack |
992 |
1436 |
145 |
16 |
129 |
| ArtL |
554 |
694 |
128 |
32 |
115 |
| Jadeplant |
994 |
1318 |
320 |
11 |
298 |
| Motorcycle |
994 |
1482 |
140 |
15 |
129 |
| MotorcycleE |
994 |
1482 |
140 |
15 |
129 |
| Piano |
962 |
1414 |
130 |
18 |
109 |
| PianoL |
962 |
1414 |
130 |
18 |
109 |
| Pipes |
970 |
1470 |
150 |
13 |
138 |
| Playroom |
952 |
1398 |
165 |
15 |
151 |
| Playtable |
926 |
1360 |
145 |
13 |
136 |
| PlaytableP |
924 |
1362 |
145 |
13 |
135 |
| Recycle |
972 |
1440 |
130 |
16 |
114 |
| Shelves |
994 |
1476 |
120 |
26 |
103 |
| Teddy |
750 |
900 |
128 |
25 |
110 |
| Vintage |
960 |
1444 |
380 |
17 |
362 |
ETH3D Info:
| Images (including 27 images) |
Height |
Width |
Max Disparity |
| delivery_area_1s |
435 |
711 |
711 |
| electro_1l |
489 |
927 |
927 |
| electro_2s |
435 |
717 |
717 |
| facade_1s |
425 |
707 |
707 |
| forest_1s |
441 |
715 |
715 |
| forest_2s |
440 |
715 |
715 |
| playground_1l |
490 |
941 |
941 |
| playground_1s |
436 |
712 |
712 |
| playground_2l |
490 |
941 |
941 |
| playground_3l |
490 |
941 |
941 |
| playground_3s |
436 |
712 |
712 |
| terrains_1l |
491 |
940 |
940 |
| terrains_1s |
438 |
713 |
713 |
| terrains_2l |
491 |
940 |
940 |
| terrains_2s |
438 |
713 |
713 |
| playground_2s |
436 |
712 |
712 |
| delivery_area_3l |
489 |
942 |
942 |
| delivery_area_1l |
489 |
942 |
942 |
| delivery_area_3s |
435 |
711 |
711 |
| terrace_1s |
434 |
713 |
713 |
| terrace_2s |
434 |
713 |
713 |
| electro_2l |
489 |
927 |
927 |
| delivery_area_2l |
489 |
942 |
942 |
| delivery_area_2s |
435 |
711 |
711 |
| electro_1s |
435 |
717 |
717 |
| electro_3s |
435 |
717 |
717 |
| electro_3l |
489 |
927 |
927 |