Devon: Deformable Volume Network for Learning Optical Flow

TL;DR

Devon introduces a deformable volume network that improves optical flow estimation, especially for fast-moving small objects, by addressing multi-resolution and warping artifacts in existing methods.

Contribution

The paper presents a novel Deformable Cost Volume module and Devon network that estimate multi-scale optical flow in high resolution, overcoming limitations of traditional multi-resolution warping approaches.

Findings

Better handling of small, fast-moving objects.

Achieves comparable accuracy to state-of-the-art methods.

Reduces artifacts caused by occlusion and dis-occlusion.

Abstract

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.