-
For all datasets, the results are reported on the test set.
-
The inference time is averaged over 100 runs, with batch size 1 on a single NVIDIA V100 GPU.
For Scene Flow dataset, the inference resolution is 576x960. For KITTI dataset, the resolution is 384x1248.
| Method | EPE | > 1px | Params | Time(ms) | Download |
|---|---|---|---|---|---|
| StereoNet-AA | 1.08 | 12.9 | 0.53M | 17 | model |
| GC-Net-AA | 0.98 | 10.8 | 2.15M | 91 | model |
| PSMNet-AA | 0.97 | 10.2 | 4.15M | 77 | model |
| GA-Net-AA | 0.87 | 9.2 | 3.68M | 57 | model |
| AANet | 0.87 | 9.3 | 3.93M | 68 | model |
| AANet+ | 0.72 | 7.4 | 8.44M | 64 | model |
| Method | Out-Noc | Out-All | Params | Time(ms) | Download |
|---|---|---|---|---|---|
| AANet | 1.91 | 2.42 | 3.93M | 62 | model |
| AANet+ | 1.55 | 2.04 | 8.44M | 60 | model |
| Method | D1-bg | D1-all | Params | Time(ms) | Download |
|---|---|---|---|---|---|
| AANet | 1.99 | 2.55 | 3.93M | 62 | model |
| AANet+ | 1.65 | 2.03 | 8.44M | 60 | model |