A mask-conditioned background loss and occlusion-aware pruning produce compact object-centric 2D Gaussian Splatting models that are up to 96% smaller and 71% faster to train.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Object-Centric 2D Gaussian Splatting: Background Removal and Occlusion-Aware Pruning for Compact Object Models
A mask-conditioned background loss and occlusion-aware pruning produce compact object-centric 2D Gaussian Splatting models that are up to 96% smaller and 71% faster to train.