SMPL Normal Map Is All You Need for Single-view Textured Human Reconstruction

2025-06-15

AI Summary

Single-view textured human reconstruction aims to recover a textured 3D human mesh from a single RGB image; however, existing feed-forward methods are hampered by the scarcity of 3D ground-truth annotations, while diffusion-based approaches often introduce 2D artifacts and geometric inconsistencies. This paper proposes a novel feed-forward framework guided and constrained by surface normal maps—eliminating the need for diffusion models and enabling geometry-consistent, texture-complete SMPL mesh generation in a single forward pass. Our method jointly leverages priors from large-scale 3D reconstruction models, SMPL parameterized human shape, a normal-map auxiliary network, and a normal-space Gaussian distribution prediction constraint. Evaluated on two mainstream benchmarks, our approach significantly outperforms state-of-the-art methods: it improves texture fidelity in visible regions and enhances geometric plausibility in occluded areas. Results demonstrate that explicit modeling in normal space substantially boosts both effectiveness and robustness for single-view human reconstruction.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Learning with Manifolds
• Natural Language Processing: Generation

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Single-view textured human reconstruction aims to reconstruct a clothed 3D digital human by inputting a monocular 2D image. Existing approaches include feed-forward methods, limited by scarce 3D human data, and diffusion-based methods, prone to erroneous 2D hallucinations. To address these issues, we propose a novel SMPL normal map Equipped 3D Human Reconstruction (SEHR) framework, integrating a pretrained large 3D reconstruction model with human geometry prior. SEHR performs single-view human reconstruction without using a preset diffusion model in one forward propagation. Concretely, SEHR consists of two key components: SMPL Normal Map Guidance (SNMG) and SMPL Normal Map Constraint (SNMC). SNMG incorporates SMPL normal maps into an auxiliary network to provide improved body shape guidance. SNMC enhances invisible body parts by constraining the model to predict an extra SMPL normal Gaussians. Extensive experiments on two benchmark datasets demonstrate that SEHR outperforms existing state-of-the-art methods.

Problem

• Reconstruct 3D clothed humans from single 2D images
• Overcome limitations of scarce 3D data and 2D hallucinations
• Enhance accuracy using SMPL normal map guidance and constraints

Innovation

• Uses SMPL normal map guidance
• Integrates pretrained 3D reconstruction model
• Predicts extra SMPL normal Gaussians