AViD — Dataset for Recognizing Video Actions

AViD is a public dataset with anonymized videos from different countries. Dataset is intended for the task of recognizing actions. AViD consists of a video where a person performs one action. The diversity in the dataset is provided by the fact that the videos were collected from different countries. The main purpose of AViD is to provide the ability to train generic models of action recognition, rather than country-specific ones. All faces in the videos have been anonymized to protect privacy. Data is licensed under the Creative Commons license.