InvariantOODG | Zhimin Zhang

InvariantOODG addresses the domain gap between synthetic and real 3D point clouds for out-of-distribution generalization.

The method uses a two-branch network to extract local-to-global features from original and augmented point clouds. A set of learnable anchor points identifies useful local regions, and two augmentation transformations improve invariant representation learning.

Experiments on 3D domain generalization benchmarks show strong robustness on unseen scenes.

(Zhang et al., 2024)

References

2024

ICASSP

InvariantOODG: Learning Invariant Features of Point Clouds for Out-of-Distribution Generalization

Zhimin Zhang, Xiang Gao, and Wei Hu

In IEEE International Conference on Acoustics, Speech and Signal Processing, 2024

Bib

@inproceedings{zhang2024invariantoodg,
  title = {InvariantOODG: Learning Invariant Features of Point Clouds for Out-of-Distribution Generalization},
  author = {Zhang, Zhimin and Gao, Xiang and Hu, Wei},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2024},
  pages = {7420--7424},
  archiveprefix = {arXiv},
  url = {https://arxiv.org/abs/2401.03765},
}