DeformationBasis

Official code and model release for our NeurIPS 2022 paper Reduced Representation of Deformation Fields for Effective Non-rigid Shape Matching.

Setup

This code is primarily written in Python 3.7 using Pytorch but with additional dependencies that are not packaged using pip/conda.

1. Setup the conda environment with Pytorch, Trimesh, etc.. from my_env.yml.

2. Install Chamfer's Distance from here.

3. Install pymesh as follows,

pip install http://imagine.enpc.fr/~langloip/data/pymesh2-0.2.1-cp37-cp37m-linux_x86_64.whl

4. Setup respective dataset, ground truth directories in local_config.py. Please follow the instructions in the comments provided.

5. For evaluation, we use CorrespondenceEvaluator package. To setup please do as follows,

git clone https://github.com/Sentient07/CorrespondenceEvaluator.git && cd CorrespondenceEvaluator && pip install -e .

Training

We provide the template data including pre-computed basis function, nodes, etc in ./data/ directory. To train the model, run the following command,

python lit_train_MLS.py --exp_name test --id 1 --pe_enc --cd_w_volp --cd_w_arap

Please refer to arguments in utils/argument_parsers.py for more details.

Evaluation

Once trained use --only_test and --model arguments to restore the model and evaluate it, e.g.,

python lit_train_MLS.py --exp_name test --id 1 --pe_enc --cd_w_volp --cd_w_arap --only_test --model /path/to/checkpoint.ckpt

Pretrained models

Pre-trained models used in our quantitative experiments can be found here: https://nuage.lix.polytechnique.fr/index.php/s/oP2zBQy7ScHRxN6

⏳ Coming Soon...

• Pre-trained weights to reproduce.

• Dataset used in all our experiments.

• Pre-processing code to obtain basis function $\Phi$, its gradient, etc.. (Check batched_mls_function.py)

• Code for shape interpolation.

Citation

If you find this code useful, please cite our paper,

@article{Sundararaman2022DeformBasis,
  title={Reduced Representation of Deformation Fields for Effective Non-rigid Shape Matching},
  author={Sundararaman, Ramana and Marin, Riccardo and Rodola, Emanuele and Ovsjanikov, Maks},
  year={2022},
  journal={Advances in Neural Information Processing Systems},
 	volume={35},
}