FlowCaps: Optical Flow Estimation with Capsule Networks For Action Recognition

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

Paper

Similar papers

© 2026 NYSGPT2525 LLC