3D Gaussian splatting
PulseAugur coverage of 3D Gaussian splatting — every cluster mentioning 3D Gaussian splatting across labs, papers, and developer communities, ranked by signal.
- 2026-05-15 research_milestone Publication of a research paper introducing a new method for robust prior-guided segmentation for editable 3D Gaussian Splatting. 来源
7 天有情绪数据
-
新的4D模型通过几何约束增强机器人操作能力
研究人员通过增强具有几何理解能力的视频世界模型，开发了新的机器人操作方法。GEM-4D将4D对应监督注入生成模型，以确保运动一致性和物理基础，将实际操作成功率从61%提高到81%。另外，GAF使用高斯动作场来表示4D中的动态场景，从而能够直接从感知运动的表示中进行动作推理，并将操作成功率提高了7.3%。这两种方法都旨在弥合逼真视频生成与可靠机器人任务执行之间的差距。
-
竹马创新获数百万融资，空间相机助力具身智能
消费级空间相机公司竹马创新（Zhuma Innovation）已完成由证大数智（SenseTime Guoxiang Capital）领投、高瓴创投（CDH VGC）和峰瑞资本（Fengrui Capital）跟投的数千万元天使+轮融资。公司旨在开创AI硬件新类别——空间相机，它既是3D内容创作工具，也是具身智能和世界模型所需真实世界3D数据的入口。其产品Pebble采用3D高斯溅射（3D Gaussian Splatting）和多传…
-
GlowGS 改进了夜景的3D高斯泼溅效果
研究人员开发了 GlowGS，这是一种用于改进夜景中3D高斯泼溅（3DGS）的新方法，尤其是在有发光效果的区域。现有的 3DGS 方法在低光照条件下由于缺乏纹理和边缘等结构特征而面临挑战。GlowGS 通过使用扩散模型和视觉基础模型（VFM）来生成和学习语义特征，从而弥补了视觉线索的缺失，实现了更准确的 3D 场景重建。
-
New CAdam method optimizes 3D Gaussian Splatting for generative distillation
Researchers have developed CAdam, a new framework to improve the efficiency of 3D Gaussian Splatting in generative distillation. This method addresses the "Densification Dilemma" by using gradient moments to distinguish…
-
RT-Splatting improves 3D rendering of transparent and reflective surfaces
Researchers have introduced RT-Splatting, a novel framework designed to improve the rendering of semi-transparent and specular surfaces in 3D scenes. This method disentangles geometric occupancy from optical opacity for…
-
LiteLoc system slashes visual localization storage and computation
Researchers have developed LiteLoc, a more efficient visual localization system that significantly reduces storage and computational demands compared to previous methods. The system achieves this by decoupling feature a…
-
New research advances 3D Gaussian Splatting for reconstruction and watermarking
Researchers have introduced several advancements in 3D Gaussian Splatting (3DGS) technology. New methods like TWINGS improve initialization for sparse-view reconstructions, enhancing detail preservation. Others, such as…
-
Learn2Splat 优化器提升 3D Gaussian Splatting 效率
研究人员开发了一种新颖的 3D Gaussian Splatting (3DGS) 学习优化器，可提高优化效率和收敛速度。这种名为 Learn2Splat 的新方法通过预测与场景结构和空间关系相关的参数更新，解决了标准优化器的局限性。它无需手动学习率调度或受限于固定迭代次数即可实现这一点，在扩展优化范围方面表现稳定，并能泛化到未见的重建场景。
-
GeoQuery framework enhances 3D reconstruction with geometry-guided diffusion
Researchers have developed GeoQuery, a new framework for 3D reconstruction that improves accuracy in sparse-view scenarios. This method integrates generative diffusion models with explicit geometric cues, overcoming lim…
-
PoseCompass improves synthetic data selection for visual localization
Researchers have developed PoseCompass, a new pipeline designed to improve the quality and efficiency of synthetic data used for training visual localization models. This system intelligently selects synthetic camera po…
-
Neural scenes enable unified radio simulation and view synthesis
Researchers have developed a novel framework that integrates differentiable ray tracing with Gaussian primitives, enabling unified simulation of radio propagation and view synthesis within neural scenes. This approach a…
-
KFC-W model generates 3D-consistent videos from unposed internet photos
Researchers have developed KFC-W, a novel self-supervised method for generating 3D-consistent videos from unposed internet photos. This approach leverages the inherent consistency of videos and the variability of multi-…
-
OpenGaFF framework enhances 3D scene understanding with Gaussian features and codebook attention
Researchers have introduced OpenGaFF, a new framework designed to improve open-vocabulary 3D scene understanding using 3D Gaussian Splatting. The system models semantics as a continuous function of Gaussian geometry and…
-
LTGS models long-term scene changes from sparse camera views
Researchers have developed LTGS, a novel method for creating long-term chronological scene representations from sparse visual updates. This approach uses 3D Gaussian Splatting to model scene changes, even with limited a…
-
New ULF-Loc method improves visual localization using unbiased landmark features with 3D Gaussian Splatting.
Researchers have developed ULF-Loc, a new framework for visual localization that addresses biases in 3D Gaussian Splatting (3DGS) features. By analyzing the $\alpha$-blending optimization in 3DGS, they identified an inh…
-
Chorus 预训练框架学习整体式 3D 高斯场景编码
研究人员开发了 Chorus，一个旨在增强 3D 高斯泼溅（3DGS）场景编码的新型预训练框架。该方法利用多个 2D 基础模型作为教师来提炼多样化信号，从而实现整体式场景表示，捕捉从高层到细粒度的语义细节。Chorus 在各种下游任务中表现出色，包括分割和数据高效监督，并且即使在适应点云基准测试时也显示出惊人的可迁移性。
-
VRGaussianAvatar integrates 3D Gaussian avatars into VR for real-time immersion
Researchers have developed VRGaussianAvatar, a system that integrates 3D Gaussian Splatting avatars into virtual reality experiences using only head-mounted display tracking. The system employs a parallel pipeline with …
-
SaLF offers real-time multi-sensor simulation for autonomous driving
Researchers have developed SaLF, a novel volumetric representation for real-time multi-sensor simulation. This method uses a sparse set of 3D voxel primitives, each containing a local implicit field, to achieve faster t…
-
World Labs releases Spark 2.0 for web-based 3DGS streaming
World Labs has released Spark 2.0, a new system for streaming Level of Detail (LoD) in 3D Gaussian Splatting (3DGS) for web applications. Built upon the Three.js framework, this open-source system aims to improve the ef…
-
GOR-IS framework improves 3D object removal with intrinsic space inpainting
Researchers have developed GOR-IS, a new framework for removing objects from 3D scene reconstructions generated by methods like 3D Gaussian Splatting. This approach addresses limitations in existing techniques by explic…