Шарафян Давид Вагеевич
Human Pose Estimation With Deep Structured Models
Прикладная математика и информатика
Articulated human pose estimation is an important topic in computer vision with many applications including human tracking and action recognition. Convolutional neural networks have achieved a remarkable performance boost in this area. Currently, in most research works, greedy algorithms are used to obtain a pose from neural network's output. In this paper, we are trying to further improve the performance of the neural network in terms of mean average precision by using the structured model as a final step of prediction. The motivation behind this is to use the information about the graphical structure of the human pose. To compare methods we use COCO 2014 keypoints challenge dataset for training and evaluation.
Текст работы (работа добавлена 19 мая 2019г.)