papers AI Learner
The Github is limit! Click to go to the new site.

Action Recognition from Single Timestamp Supervision in Untrimmed Videos

2019-04-09
Davide Moltisanti, Sanja Fidler, Dima Damen

Abstract

Recognising actions in videos relies on labelled supervision during training, typically the start and end times of each action instance. This supervision is not only subjective, but also expensive to acquire. Weak video-level supervision has been successfully exploited for recognition in untrimmed videos, however it is challenged when the number of different actions in training videos increases. We propose a method that is supervised by single timestamps located around each action instance, in untrimmed videos. We replace expensive action bounds with sampling distributions initialised from these timestamps. We then use the classifier’s response to iteratively update the sampling distributions. We demonstrate that these distributions converge to the location and extent of discriminative action segments. We evaluate our method on three datasets for fine-grained recognition, with increasing number of different actions per video, and show that single timestamps offer a reasonable compromise between recognition performance and labelling effort, performing comparably to full temporal supervision. Our update method improves top-1 test accuracy by up to 5.4%. across the evaluated datasets.

Abstract (translated by Google)
URL

http://arxiv.org/abs/1904.04689

PDF

http://arxiv.org/pdf/1904.04689


Similar Posts

Comments