ARLNC 2013 - The First International Workshop on Action Recognition with Large Number of Classes

For action recognition to operate in realistic conditions, the vision community needs to make a concerted effort to go beyond datasets with limited number of action classes, such as KTH, Weizmann and IXMAS. The goal of our workshop is to encourage researchers to develop novel methods for action recognition that scale to large numbers of action categories captured in natural settings, both in terms of classification accuracy and computational complexity.
To enable direct comparisons of proposed approaches, we will encourage workshop participants to evaluate their methods on the newly released UCF101 dataset, which is currently the largest action dataset both in terms of number of categories and clips, with more than 13000 clips drawn from 101 action classes.
Since UCF101 dataset contains more than two million frames, we recognize that computing features may itself be a challenge for those workshop participants who lack access to cluster computing resources. Therefore, in order to encourage broad participation, we made available a variety of pre-computed low-level features, such as STIP, SIFT and DTF (Dense Trajectory Feature). While participants are encouraged to employ their own features, the provided features may serve as a useful resource, particularly for computationally-constrained participants. In addition, we will make frame-by-frame bounding box annotations for humans in 25 action classes as well as class-level attribute lists.