Single-shot 3D multi-person pose estimation in complex images

In this paper, we propose a new single shot method for multi-person 3D human pose estimation in complex images. The model jointly learns to locate the joints image, estimate their coordinates and group these predictions into full skeletons. proposed deals with variable number of people does not need bounding boxes poses. It leverages extends Stacked Hourglass Network its multi-scale feature learning manage situations. Thus, exploit robust formulation fully describe several poses even case strong occlusions or crops. Then, joint grouping an arbitrary are performed using associative embedding method. Our approach significantly outperforms state art on challenging CMU Panoptic previous MuPoTS-3D dataset. Furthermore, it leads good results synthetic images from newly JTA Dataset.