FlowCaps: Optical Flow Estimation with Capsule Networks For Action Recognition

TL;DR

FlowCaps introduces a capsule network-based architecture for optical flow estimation, aiming to improve accuracy, interpretability, generalization, and computational efficiency over traditional CNN methods.

Contribution

The paper presents a novel capsule network architecture, FlowCaps, specifically designed for optical flow estimation, emphasizing better correspondence, interpretability, and efficiency.

Findings

Achieves finer-grained, motion-specific encoding.

Requires less ground truth data.

Reduces computational complexity.

Abstract

Capsule networks (CapsNets) have recently shown promise to excel in most computer vision tasks, especially pertaining to scene understanding. In this paper, we explore CapsNet's capabilities in optical flow estimation, a task at which convolutional neural networks (CNNs) have already outperformed other approaches. We propose a CapsNet-based architecture, termed FlowCaps, which attempts to a) achieve better correspondence matching via finer-grained, motion-specific, and more-interpretable encoding crucial for optical flow estimation, b) perform better-generalizable optical flow estimation, c) utilize lesser ground truth data, and d) significantly reduce the computational complexity in achieving good performance, in comparison to its CNN-counterparts.