SpatialLabFollow the Journey
RESEARCH / BUILD / EXPLORE
I explore the intersection of AI, spatial computing, and human experience — building more intelligent, immersive, and connected worlds.
SPACE
WORLDS
PEOPLE
IDEAS
POSSIBILITIES
MAP
MODEL
SIMULATE
CREATE
SPATIAL DATA
REAL WORLDS
VIRTUAL WORLDS
BRIGHTER FUTURES
BUILD
EXPLORE
UNDERSTAND
TOGETHER
Building Immersive Worlds with Spatial Intelligence — 以空间智能，构建沉浸世界。
SpatialLab is an independent research project working through spatial intelligence layer by layer: photorealistic 3D vision, Gaussian Splatting, neural rendering, and world models — from real rooms toward interactive worlds.
Explore the ProjectCurrent stage: architecture foundation. Research plans and protocols are defined; no experimental results yet.
Spatial Intelligence From Scratch · SPATIALLAB
从一条射线出发，逐步构建三维 Density/Color Field，推导局部光传输、Transmittance 与 Volume Rendering Equation，并使用 PyTorch 从零实现单射线渲染、相机模型和完整图像渲染器。最后通过 Autograd 验证像素误差如何反向更新三维场参数，建立从三维表示到图像、再从图像反馈到三维表示的可微渲染闭环。
Read More ：从一个像素开始：从零构建可微 Volume RendererSpatial Intelligence From Scratch · SPATIALLAB
从上一篇可微 Volume Renderer 出发，使用已知相机参数的多视图 RGB 图像，从零构建一个完整的最小 NeRF。逐步实现 Positional Encoding、Stratified Sampling、Hierarchical Sampling 与 Coarse/Fine 训练，最终生成未见相机视角，并通过 RGB、Depth 和 Density 分析理解新视角合成与真实几何重建的区别。
Read More ：从零构建 NeRF：从多视图图像学习一个可渲染的三维场景Spatial Intelligence From Scratch · SPATIALLAB
承接 NeRF 篇的可微渲染与多视图场景学习，将隐式 Radiance Field 转换为可学习的显式 3D Gaussian 表示。逐步实现 Gaussian Shape、Perspective Projection、Alpha Compositing 和图像监督优化，并引入 SfM 初始化、Adaptive Density Control、Spherical Harmonics 与 Tile-Based Rasterization。最终完成新视角渲染与场景保存，通过 RGB、Depth、Geometry 和性能诊断理解高质量图像生成与准确三维重建的区别，为下一篇 RoomGS 真实室内重建奠定基础。
Read More ：从零构建 3D Gaussian Splatting：用显式高斯学习一个可渲染的三维场景
LiDAR, photogrammetry, scene capture, open datasets
BROWSE THE NOTES ：Spatial Data & 3D Models
3D rendering, spatial AI, visual understanding
BROWSE THE NOTES ：Analysis & Visualization
Open source tools and reproducible workflows
BROWSE THE NOTES ：Open Infrastructure
CONCEPTExploring urban reconstruction and interactive 3D environments.
CONCEPTExploring natural environments through immersive visualization.
CONCEPTReconstructing and exploring interior spaces in 3D.
SAME PLANET
MORE
PERSPECTIVE
Follow the research notes, experiments, and ideas as the project unfolds. Email updates are not open yet — RSS is the best way to stay current.
Follow via RSSEXPLORE
LEARN
BUILD
BELONG
