Devon: Deformable Volume Network for Learning Optical Flow
About
State-of-the-art neural network models estimate large displacement optical flow in multi-resolution and use warping to propagate the estimation between two resolutions. Despite their impressive results, it is known that there are two problems with the approach. First, the multi-resolution estimation of optical flow fails in situations where small objects move fast. Second, warping creates artifacts when occlusion or dis-occlusion happens. In this paper, we propose a new neural network module, Deformable Cost Volume, which alleviates the two problems. Based on this module, we designed the Deformable Volume Network (Devon) which can estimate multi-scale optical flow in a single high resolution. Experiments show Devon is more suitable in handling small objects moving fast and achieves comparable results to the state-of-the-art methods in public benchmarks.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Optical Flow Estimation | KITTI 2015 (train) | Fl-epe2 | 431 | |
| Optical Flow Estimation | MPI Sintel Final (train) | Endpoint Error (EPE)2.67 | 209 | |
| Optical Flow Estimation | MPI Sintel Clean (train) | EPE1.97 | 202 | |
| Optical Flow | MPI Sintel Clean (test) | AEE4.34 | 158 | |
| Optical Flow | KITTI 2012 (train) | AEE1.29 | 115 | |
| Optical Flow Estimation | Sintel clean (test) | EPE4.34 | 103 | |
| Optical Flow Estimation | Sintel Final (test) | EPE6.35 | 101 | |
| Optical Flow | KITTI-15 (test) | Fl-all14.31 | 85 | |
| Optical Flow Estimation | KITTI flow 2012 (test) | EPE2.6 | 24 |