With the emergence of Gaussian Splats, recent efforts have focused on
large-scale scene geometric reconstruction. However, most of these efforts
either concentrate on memory reduction or spatial space division, neglecting
information in the semantic space. In this paper, we propose a novel method,
named SA-GS, for fine-grained 3D geometry reconstruction using semantic-aware
3D Gaussian Splats. Specifically, we leverage prior information stored in large
vision models such as SAM and DINO to generate semantic masks. We then
introduce a geometric complexity measurement function to serve as soft
regularization, guiding the shape of each Gaussian Splat within specific
semantic areas. Additionally, we present a method that estimates the expected
number of Gaussian Splats in different semantic areas, effectively providing a
lower bound for Gaussian Splats in these areas. Subsequently, we extract the
point cloud using a novel probability density-based extraction method,
transforming Gaussian Splats into a point cloud crucial for downstream tasks.
Our method also offers the potential for detailed semantic inquiries while
maintaining high image-based reconstruction results. We provide extensive
experiments on publicly available large-scale scene reconstruction datasets
with highly accurate point clouds as ground truth and our novel dataset. Our
results demonstrate the superiority of our method over current state-of-the-art
Gaussian Splats reconstruction methods by a significant margin in terms of
geometric-based measurement metrics. Code and additional results will soon be
available on our project page.

采用语义感知的 3D 高斯斑点，提出了一种新的方法 SA-GS 用于细粒度的 3D 几何重建，并在大规模场景重建数据集上展示了其卓越性能。

SA-GS：基于语义感知的高斯点渲染用于带有几何约束的大场景重构

SA-GS: Semantic-Aware Gaussian Splatting for Large Scene Reconstruction  with Geometry Constrain

We propose to formulate point cloud extraction from ultrasound volumes as an
image segmentation problem. Through this convenient formulation, a quick
prototype exploring various variants of the U-Net architecture was developed
and evaluated. This report documents the experimental results compiled using a
training dataset of 5 labelled ultrasound volumes and 84 unlabelled volumes
that got completed in a two-week period as part of a challenge submission to an
open challenge entitled ``Deep Learning in Ultrasound Image Analysis''. Source
code is shared with the research community at this GitHub URL
https://github.com/lisatwyw/smrvis.

本文提出将超声波成像中的点云提取问题转化为图像分割问题，并通过使用 U-Net 架构的不同变体进行快速原型制作和评估来解决问题。作者使用包含 5 个标记超声波成像体积和 84 个未标记体积的训练数据集，在两周内完成了实验，并在开放挑战赛 “超声图像分析中的深度学习” 中提交了挑战结果进行比较分析。作者共享了源代码，以供研究社区使用。