Generating 3D People in Scenes Without People

A Review of Body Measurement Using 3D Scanning

“…The purpose of PCA is to compress the dataset of registered meshes by finding pose and shape principal components that explain the maximal variance of the dataset. An important advantage of PCA is that the PCs can be used to generate novel template meshes [107], [122], [149], [177] from a pose-shape parameter space. The datasets commonly used for building SMs are CAESAR [3], Size-UK [8], ScanDB [64], and possibly other datasets containing 3D scans [12], [24], [72], [148], [175].…”

Grasping Field: Learning Implicit Representations for Human Grasps

“…Furthermore, in contrast to previous works, e.g. [26,29,94], which can only evaluate body-object interactions after obtaining the body and the object meshes, when using GF as the representation in hand-object reconstruction from images, we model the hand, the object, and the contact area by the implicit surfaces in a common space, largely improving the physical plausibility of the reconstruction.…”

PLACE: Proximity Learning of Articulation and Contact in 3D Environments

“…To achieve high realism, manual animation of human-scene interactions is often required. Recent works [21,42] have proposed methods for the fully automated generation of human bodies that interact with the 3D world, and yet the naturalness and the realism of their results are still far behind the captured real human-scene interaction such as [14].…”

“…To explicitly model the contact relationship between the body and the scene is key to the realism of the synthesized humans in the scene. However, existing methods use images, semantic and depth maps to represent the 3D environment [21,42]. While being easily integrated into deep convolutional networks, the 3D scene structure and the proximity between the body and the scene are not explicitly modeled, especially for the regions that are occluded from the camera view, making it hard to effectively enforce constraints in 3D, such as no inter-penetration and proper contact.…”

“…We further infer a plausible full body mesh that complies with the generated human-scene proximity. Following [14,42], we refine the body mesh via optimization; going beyond prior work, we introduce a novel objective term based on our interaction representation that further improve the performance.…”