Introduction

This repository is the official implementation of GarmentDiffusion, a generative model for 3D sewing pattern generation.

The key features of our model are:

• Multimodal Inputs: Our model takes in text, image, and incomplete sewing pattern as input modalities.
• Efficient Edge Encoding Scheme: We use an edge-oriented encoding method to encode the 3D sewing pattern parameters into compact edge token representations. This significantly reduces the sequence length of the edge tokens, allowing for faster training and generation.
• Diffusion Transformer: Our model uses a diffusion transformer to denoise all edge tokens along the temporal axis, maintaining a constant number of denoising steps regardless of dataset-specific edge and panel statistics.

We evaluate our model thoroughly on the DressCodeData, SewFactory and GarmentCodeData to validate the effectiveness of our method.

Acknowledgment

This project takes the following works ([BRepGen], [DressCode], [PyGarment], [SewFormer]) as reference. We thank the authors of above projects for their great works.

We especially thank the authors of ([Garment-Pattern-Generator], [GarmentCodeData]) for providing the great garment sewing pattern datasets.

Citation

@inproceedings{ijcai2025p163,
  title     = {GarmentDiffusion: 3D Garment Sewing Pattern Generation with Multimodal Diffusion Transformers},
  author    = {Li, Xinyu and Yao, Qi and Wang, Yuanda},
  booktitle = {Proceedings of the Thirty-Fourth International Joint Conference on
               Artificial Intelligence, {IJCAI-25}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {James Kwok},
  pages     = {1458--1466},
  year      = {2025},
  month     = {8},
  note      = {Main Track},
  doi       = {10.24963/ijcai.2025/163},
  url       = {https://doi.org/10.24963/ijcai.2025/163},
}