Efficient Semantic Splatting for Remote Sensing Multi-view Segmentation

TL;DR

This paper introduces an efficient semantic splatting method using Gaussian Splatting for fast, low-latency multi-view segmentation of point clouds, combining explicit structure and novel pseudo-labeling techniques.

Contribution

It presents a new semantic splatting approach that enhances efficiency and accuracy in multi-view segmentation by leveraging Gaussian Splatting and boundary-aware pseudo-labeling.

Findings

Achieves faster rendering and optimization times.

Improves segmentation accuracy at object boundaries.

Ensures view consistency and spatial continuity.

Abstract

In this paper, we propose a novel semantic splatting approach based on Gaussian Splatting to achieve efficient and low-latency. Our method projects the RGB attributes and semantic features of point clouds onto the image plane, simultaneously rendering RGB images and semantic segmentation results. Leveraging the explicit structure of point clouds and a one-time rendering strategy, our approach significantly enhances efficiency during optimization and rendering. Additionally, we employ SAM2 to generate pseudo-labels for boundary regions, which often lack sufficient supervision, and introduce two-level aggregation losses at the 2D feature map and 3D spatial levels to improve the view-consistent and spatial continuity.