Learning Affinity with Hyperbolic Representation for Spatial Propagation

Jin-Hwi Park, Jaesung Choe, Inhwan Bae, Hae-Gon Jeon
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:27048-27073, 2023.

Abstract

Recent approaches to representation learning have successfully demonstrated the benefits in hyperbolic space, driven by an excellent ability to make hierarchical relationships. In this work, we demonstrate that the properties of hyperbolic geometry serve as a valuable alternative to learning hierarchical affinity for spatial propagation tasks. We propose a Hyperbolic Affinity learning Module (HAM) to learn spatial affinity by considering geodesic distance on the hyperbolic space. By simply incorporating our HAM into conventional spatial propagation tasks, we validate its effectiveness, capturing the pixel hierarchy of affinity maps in hyperbolic space. The proposed methodology can lead to performance improvements in explicit propagation processes such as depth completion and semantic segmentation.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-park23b, title = {Learning Affinity with Hyperbolic Representation for Spatial Propagation}, author = {Park, Jin-Hwi and Choe, Jaesung and Bae, Inhwan and Jeon, Hae-Gon}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {27048--27073}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/park23b/park23b.pdf}, url = {https://proceedings.mlr.press/v202/park23b.html}, abstract = {Recent approaches to representation learning have successfully demonstrated the benefits in hyperbolic space, driven by an excellent ability to make hierarchical relationships. In this work, we demonstrate that the properties of hyperbolic geometry serve as a valuable alternative to learning hierarchical affinity for spatial propagation tasks. We propose a Hyperbolic Affinity learning Module (HAM) to learn spatial affinity by considering geodesic distance on the hyperbolic space. By simply incorporating our HAM into conventional spatial propagation tasks, we validate its effectiveness, capturing the pixel hierarchy of affinity maps in hyperbolic space. The proposed methodology can lead to performance improvements in explicit propagation processes such as depth completion and semantic segmentation.} }
Endnote
%0 Conference Paper %T Learning Affinity with Hyperbolic Representation for Spatial Propagation %A Jin-Hwi Park %A Jaesung Choe %A Inhwan Bae %A Hae-Gon Jeon %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-park23b %I PMLR %P 27048--27073 %U https://proceedings.mlr.press/v202/park23b.html %V 202 %X Recent approaches to representation learning have successfully demonstrated the benefits in hyperbolic space, driven by an excellent ability to make hierarchical relationships. In this work, we demonstrate that the properties of hyperbolic geometry serve as a valuable alternative to learning hierarchical affinity for spatial propagation tasks. We propose a Hyperbolic Affinity learning Module (HAM) to learn spatial affinity by considering geodesic distance on the hyperbolic space. By simply incorporating our HAM into conventional spatial propagation tasks, we validate its effectiveness, capturing the pixel hierarchy of affinity maps in hyperbolic space. The proposed methodology can lead to performance improvements in explicit propagation processes such as depth completion and semantic segmentation.
APA
Park, J., Choe, J., Bae, I. & Jeon, H.. (2023). Learning Affinity with Hyperbolic Representation for Spatial Propagation. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:27048-27073 Available from https://proceedings.mlr.press/v202/park23b.html.

Related Material