ICLR 2026 Orals

Generative Human Geometry Distribution

Xiangjun Tang, Biao Zhang, Peter Wonka

Diffusion & Flow Matching Thu, Apr 23 · 3:15 PM–3:25 PM · 204 A/B Avg rating: 5.50 (2–8)
Author-provided TL;DR

We introduce the first method that integrates geometry distributions into generative modeling.

Abstract

Realistic human geometry generation is an important yet challenging task, requiring both the preservation of fine clothing details and the accurate modeling of clothing-body interactions. To tackle this challenge, we build upon Geometry distributions—a recently proposed representation that can model a single human geometry with high fidelity using a flow matching model. However, extending a single-geometry distribution to a dataset is non-trivial and inefficient for large-scale learning. To address this, we propose a new geometry distribution model by two key techniques: (1) encoding distributions as 2D feature maps rather than network parameters, and (2) using SMPL models as the domain instead of Gaussian and refining the associated flow velocity field. We then design a generative framework adopting a two-staged training paradigm analogous to state-of-the-art image and 3D generative models. In the first stage, we compress geometry distributions into a latent space using a diffusion flow model; the second stage trains another flow model on this latent space. We validate our approach on two key tasks: pose-conditioned random avatar generation and avatar-consistent novel pose synthesis. Experimental results demonstrate that our method outperforms existing state-of-the-art methods, achieving a 57% improvement in geometry quality.

One-sentence summary·Auto-generated by claude-haiku-4-5-20251001(?)

Introduces distribution-over-distribution model combining geometry distributions with two-stage flow matching for human 3D generation.

Contributions·Auto-generated by claude-haiku-4-5-20251001(?)
  • Novel representation encoding geometry distributions as 2D feature maps rather than network parameters
  • Two-stage training paradigm: first compresses distributions to latent space, then learns on latent space
  • Achieves 57% improvement in geometry quality on pose-conditioned generation and novel pose synthesis
Methods used·Auto-generated by claude-haiku-4-5-20251001(?)
  • Flow matching
  • Diffusion models
  • SMPL models
  • Geometry distributions
  • Latent space learning
Limitations (author-stated)·Auto-generated by claude-haiku-4-5-20251001(?)
  • Non-uniform sampling on target geometry surface due to varying number of target points per SMPL point
    from the paper
  • Constrained by diversity of training datasets; model generalizes well to various body types but cannot generate clothing styles absent from training data
    from the paper
  • Use of UV maps can cause seam artifacts in randomly generated results due to discontinuity segmentation
    from the paper
Future work (author-stated)·Auto-generated by claude-haiku-4-5-20251001(?)

Authors did not state explicit future directions.

Author keywords

  • 3D Generation
  • Human Generation
  • Geometry Encoding

Related orals

Something off? Let us know →