Reproduce “GarmentX: Autoregressive Parametric Representations for High-Fidelity 3D Garment Generation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jingfeng Guo, Jinnan Chen, Weikai Chen, Zhenyu Sun, Lanjiong Li, Baozhu Zhao, Lingting Zhu, Xin Wang and 1 more

GarmentX introduces an autoregressive parametric representation framework for generating high-fidelity, physically valid 3D garments directly from single images. The approach produces wearable 3D models with detailed structural elements and achieves state-of-the-art performance against existing methods, including a zero simulation failure rate.

Key insights

  • A structured parametric representation of garments ensures physical validity, preserves relative positional relationships, and enables intuitive editing of attributes.
  • Combining a masked autoregressive model with diffusion effectively predicts both discrete and continuous garment parameters from image features.
  • Training on a large-scale, synthetically constructed dataset of garment parameter-image pairs allows the model to learn diverse styles and structural elements.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.