Embarrassingly Simple Model for Early Action Proposal

TL;DR

This paper introduces a simple, classifier-based 3D CNN model for real-time early action proposal in videos, outperforming more complex existing methods in online scenarios.

Contribution

The paper presents a straightforward 3D CNN approach for online action proposal, emphasizing simplicity and improved performance over complex prior methods.

Findings

Outperforms state-of-the-art methods in early action proposal

Uses standard 3D CNNs for online detection

Achieves significant improvements in real-time scenarios

Abstract

Early action proposal consists in generating high quality candidate temporal segments that are likely to contain an action in a video stream, as soon as they happen. Many sophisticated approaches have been proposed for the action proposal problem but from the off-line perspective. On the contrary, we focus on the on-line version of the problem, proposing a simple classifier-based model, using standard 3D CNNs, that performs significantly better than the state of the art.