/SewFormer

Primary LanguageJupyter Notebook

Sewformer

This is the official implementation of Towards Garment Sewing Pattern Reconstruction from a Single Image.

Lijuan Liu *, Xiangyu Xu *, Zhijie Lin *, Jiabing Liang *, Shuicheng Yan†,
ACM Transactions on Graphics (SIGGRAPH Asia 2023)

Project | Paper

Installation and Configuration

  • Clone this repository to path_to_dev and cd path_to_dev/Sewformer, download the pre-trained checkpoint and put it into assets/ckpts.
  • The environment can be initialized with conda env create -f environment.yaml. Then you can activate the environment conda activate garment.

Training

  • Download our provided dataset and put it into path_to_sewfactory, update the local paths in system.json to make sure the dataset setup correctly.

  • Train the model with torchrun --standalone --nnodes=1 --nproc_per_node=1 train.py -c configs/train.yaml

    The output will be located at the output in system.json.

Testing

  1. Inference sewing patterns with the pretrained model:

  • evaluate on sewfactory dataset: torchrun --standalone --nnodes=1 --nproc_per_node=1 train.py -c configs/train.yaml -t

  • inference on real images (e.g. from deepfashion): python inference.py -c configs/test.yaml -d assets/data/deepfashion -t deepfashion -o outputs/deepfashion

  1. Simulate the predicted results (Windows): cd path_to_dev/SewFactory and run path_to_maya\bin\mayapy.exe .\data_generator\deepfashion_sim.py to simulate the predicted sew patterns. (Please prepare the SMPL prediction results with RSC-Net and update the predicted data root specified in deepfashion_sim.py.)

    See more details about the SewFactory dataset and the simulation here.

BibTex

Please cite this paper if you find the code/model helpful in your research:

 @article{liu2023sewformer,
    author      = {Liu, Lijuan and Xu, Xiangyu and Lin, Zhijie and Liang, Jiabin and Yan, Shuicheng},
    title       = {Towards Garment Sewing Pattern Reconstruction from a Single Image},
    journal     = {ACM Transactions on Graphics (SIGGRAPH Asia)},
    year        = {2023}
  }