Actionness Estimation Using Hybrid Fully Convolutional Networks

TL;DR

This paper introduces a hybrid fully convolutional network (H-FCN) for estimating actionness in videos, combining appearance and motion cues to improve efficiency and accuracy, and demonstrates its effectiveness on multiple datasets.

Contribution

The paper proposes a novel hybrid FCN architecture that leverages static appearance and dynamic motion for improved actionness estimation in videos.

Findings

H-FCN outperforms previous methods on Stanford40, UCF Sports, and JHMDB datasets.

Actionness maps generated by H-FCN enhance action proposal and detection performance.

The method processes videos efficiently regardless of size.

Abstract

Actionness was introduced to quantify the likelihood of containing a generic action instance at a specific location. Accurate and efficient estimation of actionness is important in video analysis and may benefit other relevant tasks such as action recognition and action detection. This paper presents a new deep architecture for actionness estimation, called hybrid fully convolutional network (H-FCN), which is composed of appearance FCN (A-FCN) and motion FCN (M-FCN). These two FCNs leverage the strong capacity of deep models to estimate actionness maps from the perspectives of static appearance and dynamic motion, respectively. In addition, the fully convolutional nature of H-FCN allows it to efficiently process videos with arbitrary sizes. Experiments are conducted on the challenging datasets of Stanford40, UCF Sports, and JHMDB to verify the effectiveness of H-FCN on actionness estimation, which demonstrate that our method achieves superior performance to previous ones. Moreover, we apply the estimated actionness maps on action proposal generation and action detection. Our actionness maps advance the current state-of-the-art performance of these tasks substantially.