imGHUM: Implicit Generative Models of 3D Human Shape and Articulated Pose

We present imGHUM, the first holistic generative model of 3D human shape and\narticulated pose, represented as a signed distance function. In contrast to\nprior work, we model the full human body implicitly as a function\nzero-level-set and without the use of an explicit template mesh. We propose a\nnovel network architecture and a learning paradigm, which make it possible to\nlearn a detailed implicit generative model of human pose, shape, and semantics,\non par with state-of-the-art mesh-based models. Our model features desired\ndetail for human models, such as articulated pose including hand motion and\nfacial expressions, a broad spectrum of shape variations, and can be queried at\narbitrary resolutions and spatial locations. Additionally, our model has\nattached spatial semantics making it straightforward to establish\ncorrespondences between different shape instances, thus enabling applications\nthat are difficult to tackle using classical implicit representations. In\nextensive experiments, we demonstrate the model accuracy and its applicability\nto current research problems.\n