MrNeRF (@janusch_patas)
2024-10-24 | โค๏ธ 38 | ๐ 9
PLGS: Robust Panoptic Lifting with 3D Gaussian Splatting
Contributions (cited): โข We introduce a new framework named PLGS to lift noisy masks from 2D to 3D using 3DGS, enabling the rapid generation of consistent panoptic segmentation masks across views without ground-truth annotations.
โข To mitigate the negative effects of noisy and inconsistent mask supervision, we structure the 3D Gaussian to introduce smoothness and design effective noise reduction strategies for both semantic and instance segmentation, enhancing the robustness of our approach.
โข Extensive experiments on HyperSim, Replica and ScanNet datasets demonstrate that PLGS out- performs previous state-of-the-art methods in terms of both panoptic segmentation quality and training/rendering speed.
๋ฏธ๋์ด
Tags
domain-vision-3d domain-rendering domain-ai-ml domain-dev-tools