OAS-Net: Occlusion Aware Sampling Network for Accurate Optical Flow

TL;DR

OAS-Net introduces an occlusion-aware sampling method and a dedicated occlusion module to improve optical flow accuracy, avoiding warping artifacts and better handling occluded regions, validated on Sintel and KITTI datasets.

Contribution

The paper proposes a lightweight optical flow network with a novel sampling correlation layer and an occlusion-aware module, enhancing accuracy over traditional warping-based methods.

Findings

Outperforms existing methods on Sintel and KITTI datasets.

Effectively handles occlusions and reduces ghosting artifacts.

Achieves high accuracy with a compact network architecture.

Abstract

Optical flow estimation is an essential step for many real-world computer vision tasks. Existing deep networks have achieved satisfactory results by mostly employing a pyramidal coarse-to-fine paradigm, where a key process is to adopt warped target feature based on previous flow prediction to correlate with source feature for building 3D matching cost volume. However, the warping operation can lead to troublesome ghosting problem that results in ambiguity. Moreover, occluded areas are treated equally with non occluded regions in most existing works, which may cause performance degradation. To deal with these challenges, we propose a lightweight yet efficient optical flow network, named OAS-Net (occlusion aware sampling network) for accurate optical flow. First, a new sampling based correlation layer is employed without noisy warping operation. Second, a novel occlusion aware module is presented to make raw cost volume conscious of occluded regions. Third, a shared flow and occlusion awareness decoder is adopted for structure compactness. Experiments on Sintel and KITTI datasets demonstrate the effectiveness of proposed approaches.