Few-Shot Physically-Aware Articulated Mesh Generation via Hierarchical Deformation

ICCV 2023
Project page | Paper | Supp | Video

res-demo.mp4

[2024.10] Repository Status: Not Actively Maintained -- Thank you for stopping by. Due to limited time, this repository is currently not being maintained. I am sorry that I have been unable to complete the remaining TODOs as planned and currently have no plans to revisit, organize, and release the unfinished parts. Sharing fuzzy code privately would also make me uncomfortable. I sincerely apologize for any inconvenience this may cause and appreciate your understanding.

Enviroment

conda env create -f environment.yml

This script will create an environment named fewartigen.

Data

Please refer to doc_data for datasets and pre-processed data-related information.

Instructions

The method consists of four stages:

• Convex decomposition: a pre-processing stage. Please refer to doc_convex_decomposition for details.
• Convex deformation: learning the convex deformation module. Please refer to doc_convex_deformation for details.
• Deformation synchronization: synchronizing convex deformations for object-level deformations. Please refer to doc_deformation_synchronization for details.
• Physics-aware correction: deformation correction considering the physical validity of the generated articulated object. Please refer to doc_physics_correction for details.

TODOs (More to come, stay tuned!)

• Pre-training process
• Data and checkpoitns
• More docs

Citation

@inproceedings{liu2023fewshot,
      title={Few-Shot Physically-Aware Articulated Mesh Generation via Hierarchical Deformation},
      author={Liu, Xueyi and Wang, Bin and Wang, He and Yi, Li},
      booktitle={International Conference on Computer Vision (ICCV)},
      year={2023}
}

Contact

Please contact [email protected] if you have any questions.

Reference

Part of the code is taken from BSP-Net-Pytorch, BAE-Net, deep_cages, and DeepMetaHandles. We thank the authors for their awesome code.