HEI-Human: A Hybrid Explicit and Implicit Method for Single-View 3D Clothed Human Reconstruction

Single-view 3D clothed human reconstruction is a challenging task, not only because of the need to infer the complex global topology of human body but also due to the requirement to recover delicate surface details. In this paper, a method named HEI-Human is proposed to hybridize an explicit model and an implicit model for 3D clothed human reconstruction. In the explicit model, the SMPL model is voxelized and then integrated into a 3D hourglass network to supervise the global geometric aligned features extraction. In the implicit model, 2D aligned features are first extracted by a 2D hourglass network, and then an implicit surface function is employed to construct the occupancy field of human body using the hybrid 2D and 3D aligned features. As the explicit model and implicit model are mutually beneficial, our HEI-Human method not only generates reconstructions with plausible global topology but also recovers rich and accurate surface details. The HEI-Human is evaluated on the current largest publicly available dataset, and the experimental results demonstrate that our method outperforms the state-of-the-art methods including DeepHuman, PIFu, and GeoPIFu.