Mobile QR Code QR CODE : The Transactions of the Korean Institute of Electrical Engineers
Title Comparison of the Methods for Jointly Learning Objects and Actions Using Realtime Object Detector
Authors 홍성준(Sungjun Hong) ; 이희성(Heesung Lee)
Page pp.138-143
ISSN 1975-8359
Keywords Object-action detection; object detection; action detection; joint learning
Abstract Most of visual detection in videos are limited to focus on objects or human actions separately. In this work, changing the classification loss of well-known realtime object detector, we introduce a detection model to jointly detect object-action pairs in videos. For detecting objects-actions in videos, we present two methods to label object-action pairs, called Cartesian product (CP) and valid Cartesian product (VCP). In experiments on the A2D dataset, we compares results on detection of object-action pairs in terms of the mean average precision.