Attention Transfer from Web Images for Video Recognition

Authors:

Abstract
The Human Interaction Image (HII) dataset is a new dataset containing Web images from Commercial Search Engines (Google, Bing and Flickr). We use keyword search to collect images corresponding to four types of interactions: handshake, highfive, hug, kiss. Then we manually filter the irrelevant images. The dataset contains 2410 images with at least 550 images per interaction. The dataset can be applied, but not limited to the following research areas: interaction recognition/prediction action recognition video analysis transfer learning Please cite the following paper if you use the HII dataset in your work (papers, articles, reports, books, software, etc): J. Li, Y. Wong, Q.Zhao, M. Kankanhalli Attention Transfer from Web Images for Video Recognition ACM Multimedia, 2017. http://doi.org/10.1145/3123266.312343
Funding Information
  • National Research Foundation Prime Minister's Office Singapore