In general, the bottom-up approach runs faster than the top-down method, but in the case of, the optimization based on binary integer programming requires a lot of time, which prevents the 3D multi-person shape reconstruction task from being processed in real time. The existing multi-person shape reconstruction method in reconstructs the 3D shapes of all persons from an input image in a bottom-up manner. However, the goal of this paper, the 3D multi-person shape reconstruction problem, has been less studied than the single-person case. Additionally, the problem we address in this paper is not related to the pose refinement in. On the other hand, our goal in this paper is to obtain dense 3D shapes that can provide richer information about multiple people from an input single RGB image, that is, 3D multi-person shape reconstruction. However, this method generates 2D image coordinates of human joints, which only provide sparse information about the target human subject. proposed a pose refinement network that outputs a refined 2D pose from an input pair of an RGB image and its corresponding noisy 2D pose, which can be used for top-down 2D multi-person pose estimation. proposed a system that can detect multiple people’s interactions in a crowded sports scene in real-time. Most 3D human shape reconstruction methods focus on a single person, but real-world applications require processing multiple persons in real time. Most recent methods for single-person shape reconstruction regress the parameters of a statistical body shape model, such as the skinned multi-person linear (SMPL) model, while using a deep neural network to reconstruct a 3D shape. In recent years, 3D human shape reconstruction from a single RGB image has been actively studied as one of the challenging tasks of computer vision, but most studies address the case of a single person. The proposed network can be learned in an end-to-end manner and process images at about 37 fps to perform the 3D multi-person shape reconstruction task in real time. Moreover, our network predicts the absolute position of the root joint while reconstructing the root-relative 3D shape, which enables reconstructing the 3D shapes of multiple persons in the camera coordinate system. Our network produces output tensors divided into grid cells to reconstruct the 3D shapes of multiple persons in a single-shot manner, where each grid cell contains information about the subject. In this paper, we propose an end-to-end learning-based model for single-shot, 3D, multi-person shape reconstruction in the camera coordinate system from a single RGB image.

Although the performance of the 3D human shape reconstruction method has improved considerably in recent years, most methods focus on a single person, reconstruct a root-relative 3D shape, and rely on ground-truth information about the absolute depth to convert the reconstruction result to the camera coordinate system.

Box shot 3d shapes