Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

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.

Yao Lu, Jack Valmadre, Heng Wang, Juho Kannala, Mehrtash Harandi, Philip H. S. Torr• 2018

Related benchmarks

TaskDatasetResultRank
Optical Flow EstimationKITTI 2015 (train)
Fl-epe2
431
Optical Flow EstimationMPI Sintel Final (train)
Endpoint Error (EPE)2.67
209
Optical Flow EstimationMPI Sintel Clean (train)
EPE1.97
202
Optical FlowMPI Sintel Clean (test)
AEE4.34
158
Optical FlowKITTI 2012 (train)
AEE1.29
115
Optical Flow EstimationSintel clean (test)
EPE4.34
103
Optical Flow EstimationSintel Final (test)
EPE6.35
101
Optical FlowKITTI-15 (test)
Fl-all14.31
85
Optical Flow EstimationKITTI flow 2012 (test)
EPE2.6
24
Showing 9 of 9 rows

Other info

Follow for update