المفاهيم الأساسية
This paper introduces UOT-UPC, a novel unsupervised point cloud completion model that leverages unbalanced optimal transport maps to effectively handle class imbalance issues often present in unpaired datasets, achieving state-of-the-art results.
الإحصائيات
The proportion of some categories, e.g., 'lamp' and 'trash bin' classes, significantly differs by more than threefold between the incomplete and complete point cloud distributions in a multi-category benchmark dataset.
UOT-UPC outperforms the second-best unpaired approach, USSPA, by more than 10% in average L1 Chamfer distance scores across ten categories.
UOT-UPC achieves 6.00 and 7.30 on TV and lamp datasets, respectively, in L1 Chamfer distance, outperforming all other models, including the supervised ones.
UOT-UPC attains F0.1% score and F1% score scores of 19.55 and 76.83, respectively, surpassing all other unpaired benchmark methods.
اقتباسات
"In this paper, we introduce a novel unpaired point cloud completion model based on the unbalanced optimal transport map."
"Our model is the first attempt to leverage UOT for unpaired point cloud completion, achieving competitive or superior results on both single-category and multi-category datasets."
"In particular, our model is especially effective in scenarios with class imbalance, where the proportions of categories are different between the incomplete and complete point cloud datasets."