GARField

GARField (Garment Attached Radiance Field) adresses the data problem in deformable object manipulation through learnable, viewpoint-free data generation.

Project website

Installation

git clone git@github.com:ddonatien/GARField.git
cd GARField
pip install -r requirements.txt

Download sample data and unzip in GARField/data

Training

python run.py --mode train_render --conf confs/sock.conf --case sock_cap_sample --gpu 0

Image generation

python run.py --mode render_novel --conf confs/sock.conf --case sock_cap_sample --gpu 0  --model <MODEL_FILE>.pth

Citation

@article{delehelle2024garfield,
  title={GARField: Addressing the visual Sim-to-Real gap in garment manipulation with mesh-attached radiance fields},
  author={Delehelle, Donatien and Caldwell, Darwin G and Chen, Fei},
  journal={arXiv preprint arXiv:2410.05038},
  year={2024}
}