From 10df20336624f236e4d485b416757a019575efca Mon Sep 17 00:00:00 2001 From: kang Date: Wed, 1 Apr 2026 09:04:08 +0800 Subject: [PATCH] auto-save 2026-04-01 09:03 (~2) --- .gitignore | 1 + index.html | 995 +++++++++++++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 960 insertions(+), 36 deletions(-) diff --git a/.gitignore b/.gitignore index dd31b51..06b8039 100644 --- a/.gitignore +++ b/.gitignore @@ -1,6 +1,7 @@ node_modules/ dist/ build/ +repos/ .env .env.local .env.production diff --git a/index.html b/index.html index e139ed1..4da02ed 100644 --- a/index.html +++ b/index.html @@ -1,46 +1,969 @@ - - - 3D 场景生成开源复现 - + + +3D 场景生成 — 开源复现计划 + - -
-

3D 场景生成开源复现

-

WorldMesh 类似的开源 3D 场景生成项目调研与源码收集，待 GPU 后复现

+ -
-

概述

-

待补充研究内容...

+ +
+ + +
+ + +
+
+
Research 2026-03-27
+

+ 3D 场景生成
开源复现计划 + 3D Scene Generation
Open-Source Reproduction +

+

+ + 从 WorldMesh 出发，梳理当前最优秀的开源 3D 场景生成项目。
+ 文本/图像输入 → 可导航的 3D 世界。源码已本地保存，待 GPU 后逐一复现。 + + + From WorldMesh to the best open-source 3D scene generation projects.
+ Text/Image → Navigable 3D Worlds. All source code saved locally, ready for GPU. + +

+
+
+
7
+
开源项目Open-Source Projects
- -
-

核心发现

-

待补充...

+
+
5
+
顶会论文Top-Venue Papers
+
+
+
11K+
+
GitHub Stars
+
+ + +
+

+ 起因：WorldMesh + Origin: WorldMesh +

+

+ WorldMesh 提出了几何优先的 3D 场景生成思路，但代码未开源。我们找到了 7 个可复现的替代方案。 + The geometry-first approach that sparked this research. Code not yet released — here are 7 reproducible alternatives. +

+ +
+

WorldMesh

+
arXiv 2603.22972 · TUM · 2026-03-24 · Manuel-Andreas Schneider, Angela Dai
+

+ + 核心思路：几何优先（Geometry-First）— 文本 → 平面图 → 3D 网格支架（墙面、地面、结构）→ + 基于网格条件的图像扩散合成外观 → 3D Gaussian Splatting 输出可导航场景。 + 支持大规模多房间生成，古罗马到赛博朋克多种风格。用户偏好测试 96.2% 优于基线。 + + + Core Idea: Geometry-First — Text → Floor Plan → 3D Mesh Scaffold (walls, floors, structure) → + Mesh-Conditioned Image Diffusion for appearance → 3D Gaussian Splatting for navigable output. + Supports large-scale multi-room generation across styles (Ancient Roman to Cyberpunk). 96.2% user preference over baselines. + +

+
+ 代码未开源Code Not Released + arXiv 预印本arXiv Preprint + Mesh-Conditioned Diffusion + 3D Gaussian Splatting +
+
+
+ + +
+

+ 第一梯队 — 直接可跑 + Tier 1 — Ready to Run +

+

+ 代码完整发布，社区验证，文档齐全 + Complete code, community-verified, well-documented +

+ +
+ + +
+
#1
+ Tier 1 +

WorldGen

+
+ 1,592 Stars + Apache-2.0 + 2025-04 + 活跃维护Actively Maintained +
+

+ + 文本/图像 → 可导航 3D 场景。基于全景生成 + Gaussian Splatting，pip 安装即用。 + 支持低显存模式（10GB），是目前上手门槛最低的方案。 + + + Text/Image → navigable 3D scene. Panorama generation + Gaussian Splatting, pip install ready. + Low VRAM mode (10GB) makes it the lowest barrier option available. + +

+
+ Text / Image + → + Panorama (FLUX.1) + → + Depth Estimation + → + 3D Gaussian Splat + → + Viser 导航Viser Navigation +
+
+ GPU: 10GB (low_vram) + GPU: 24GB (标准standard) + 代码完整Code Complete + 源码已保存Saved Locally +
+
+ + +
+
#2
+ Tier 1 +

Text2Room

+
+ 1,082 Stars + ICCV 2023 + TUM +
+

+ + 该领域的奠基之作。文本 → 带纹理的 3D 房间网格。迭代生成视图、修复、对齐深度、融合网格。 + 代码干净稳定，复现性经过大量验证。 + + + The foundational work. Text → textured 3D room mesh. Iterative view generation, inpainting, depth alignment, and mesh fusion. + Clean, stable codebase with proven reproducibility. + +

+
+ Text Prompt + → + Stable Diffusion 2 + → + Depth Alignment + → + Mesh Fusion + → + Textured 3D Mesh +
+
+ GPU: 16-24GB + 代码完整Code Complete + 源码已保存Saved Locally +
+
+ + +
+
#3
+ Tier 1 +

Infinigen

+
+ 6,878 Stars + CVPR 2023 + 2024 + Princeton + BSD +
+

+ + 普林斯顿出品，程序化生成照片级室内外 3D 场景。100% 程序化，无需外部素材。 + 最成熟的项目（3,214 commits），Mac CPU 也能跑。适合生成训练数据集。 + + + Princeton's procedural generation framework. Photorealistic indoor + outdoor 3D scenes, 100% procedural, no external assets. + Most mature project (3,214 commits). CPU compatible — works on Mac without NVIDIA GPU. + +

+
+ Procedural Rules + → + Blender Generation + → + Photorealistic Render +
+
+ CPU OK / GPU Accelerated + 代码完整Code Complete + 源码已保存Saved Locally +
+
+ + +
+
#4
+ Tier 1 +

WonderWorld

+
+ 717 Stars + CVPR 2025 +
+

+ + 最接近 WorldMesh 的方案。单张图像 → 连通的可导航 3D 场景，基于 Fast Layered Gaussian Surfels (FLAGS)。 + 支持浏览器交互式导航，每个新视角 <10 秒。 + + + Closest to WorldMesh. Single image → connected navigable 3D scenes via Fast Layered Gaussian Surfels (FLAGS). + Browser-based interactive navigation, <10s per new view. + +

+
+ Single Image + → + Depth + Segmentation + → + Layered Gaussian Surfels + → + 交互式导航Interactive Navigation +
+
+ GPU: 48GB (A6000/A100) + 代码完整Code Complete + 源码已保存Saved Locally +
+
+ +
+
+ + +
+

+ 第二梯队 — 门槛稍高 + Tier 2 — Higher Barrier +

+

+ 代码可用，构建流程较复杂或硬件要求更高 + Code available, more complex setup or higher hardware requirements +

+ +
+ + +
+
#5
+ Tier 2 +

LayerPano3D

+
+ 315 Stars + SIGGRAPH 2025 +
+

+ + 文本 → 分层 360° 全景 3D 场景 + Gaussian Splatting。沉浸感最强。 + 需编译 C++ 扩展（Ceres solver、360monodepth），构建过程较复杂。 + + + Text → layered 360° panoramic 3D scene + Gaussian Splatting. Most immersive experience. + Requires C++ extension compilation (Ceres solver, 360monodepth). + +

+
+ GPU: 16-24GB + 代码完整Code Complete + 需编译 C++C++ Build Required + 源码已保存Saved Locally +
+
+ + +
+
#6
+ Tier 2 +

RealmDreamer

+
+ 297 Stars + 3DV 2025 +
+

+ + 文本 → 3D 场景（Gaussian Splatting），用户偏好测试 88-95%。 + 提供预生成输出可跳过耗时阶段，Stage 2 训练约需数小时。 + + + Text → 3D scene via Gaussian Splatting. 88-95% user preference in studies. + Pre-generated outputs available to skip slow stages. Stage 2 training takes several hours. + +

+
+ CUDA 11.8 + 代码完整Code Complete + 训练耗时数小时Hours of Training + 源码已保存Saved Locally +
+
+ + +
+
#7
+ Tier 2 +

SceneCraft

+
+ 233 Stars + NeurIPS 2024 +
+

+ + 文本 + 空间布局 → 多房间公寓 3D 场景（NeRF）。支持复杂平面图。 + 多房间能力最接近 WorldMesh，但需要 2+ GPU。 + + + Text + spatial layout → multi-room apartment 3D scenes via NeRF. Supports complex floor plans. + Closest to WorldMesh's multi-room capability, but requires 2+ GPUs. + +

+
+ GPU: 2+ 张Required + 代码部分完整Partially Complete + 源码已保存Saved Locally +
+
+ +
+
+ + +
+

+ 快速选择指南 + Quick Decision Guide +

+

+ 根据你的场景和硬件选择最合适的项目 + Pick the right project based on your scenario and hardware +

+ +
+
+
⚡
+

最快出效果Fastest Results

+

pip 安装，最低 10GB 显存，文本直接出 3D 场景pip install, 10GB VRAM minimum, text directly to 3D scene

+ WorldGen +
+
+
🎓
+

学习经典方法Learn the Fundamentals

+

ICCV 2023 奠基之作，代码干净，论文被引最多ICCV 2023 foundational work, clean code, most cited

+ Text2Room +
+
+
💻
+

没有 NVIDIA GPUNo NVIDIA GPU

+

CPU 也能跑，基于 Blender 程序化生成，Mac 可用CPU compatible, Blender-based procedural generation, works on Mac

+ Infinigen +
+
+
🌐
+

最接近 WorldMeshClosest to WorldMesh

+

交互式导航，连通多房间，但需要 48GB 显存Interactive navigation, connected multi-room, needs 48GB VRAM

+ WonderWorld +
+
+
🎬
+

360° 沉浸全景360° Immersive Panorama

+

SIGGRAPH 级别质量，360 度环绕 3D 场景SIGGRAPH-level quality, 360-degree surround 3D scenes

+ LayerPano3D +
+
+
🏢
+

多房间公寓布局Multi-Room Apartments

+

支持平面图控制，最精确的空间布局方案Floor plan controlled, most precise spatial layout

+ SceneCraft +
+
+
+ + +
+

+ 本地源码状态 + Local Source Code Status +

+

+ 所有源码已 clone 至 repos/ 目录，待 GPU 后逐一复现 + All repos cloned to repos/ — ready for GPU +

+ +
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
项目Project本地路径Local PathStars会议VenueGPU状态Status
WorldGenrepos/WorldGen/1,592Independent10-24GB已保存Saved
Text2Roomrepos/text2room/1,082ICCV 202316-24GB已保存Saved
Infinigenrepos/infinigen/6,878CVPR 23+24CPU OK已保存Saved
WonderWorldrepos/WonderWorld/717CVPR 202548GB已保存Saved
LayerPano3Drepos/LayerPano3D/315SIGGRAPH 2516-24GB已保存Saved
RealmDreamerrepos/realmdreamer/2973DV 2025CUDA 11.8已保存Saved
SceneCraftrepos/SceneCraft/233NeurIPS 242+ GPU已保存Saved
+
+
+ + +
+

+ 技术演进 + Evolution Timeline +

+

+ 3D 场景生成领域近年关键节点 + Key milestones in 3D scene generation +

+ +
+
+
2023-06
+
Text2Room (ICCV 2023)
+
首次实现文本到带纹理 3D 房间网格，奠定该方向基础First text-to-textured 3D room mesh, foundational work
+
+
+
2023-06
+
Infinigen (CVPR 2023)
+
普林斯顿程序化生成框架，无限照片级 3D 世界Princeton's procedural generation framework for infinite photorealistic worlds
+
+
+
2023-11
+
MVDiffusion (NeurIPS 2023 Spotlight)
+
多视角一致性图像生成，成为后续全景方案的基础组件Multi-view consistent image generation, foundation for panorama-based methods
+
+
+
2024-01
+
Infinigen Indoors (CVPR 2024)
+
扩展至室内场景，程序化生成家具、厨房、浴室等Extended to indoor scenes — furniture, kitchens, bathrooms
+
+
+
2024-09
+
SceneCraft (NeurIPS 2024)
+
布局引导的多房间 3D 场景生成，NeRF 输出Layout-guided multi-room 3D scene generation via NeRF
+
+
+
2024-09
+
DreamScene360 (ECCV 2024)
+
360° 全景 Gaussian Splatting 场景生成360° panoramic Gaussian Splatting scene generation
+
+
+
2025-01
+
WonderWorld (CVPR 2025)
+
单图像交互式导航 3D 场景，Fast Layered Gaussian SurfelsSingle-image interactive navigable 3D scenes, Fast Layered Gaussian Surfels
+
+
+
2025-04
+
WorldGen
+
目前最易用的文本到 3D 场景方案，10GB 显存即可Most accessible text-to-3D scene method, 10GB VRAM minimum
+
+
+
2025-07
+
LayerPano3D (SIGGRAPH 2025)
+
分层 360° 全景 + Gaussian Splatting，超沉浸体验Layered 360° panorama + Gaussian Splatting, hyper-immersive
+
+
+
2026-03
+
WorldMesh (arXiv)
+
几何优先策略，网格支架约束多房间生成，代码待发布Geometry-first mesh scaffold for multi-room generation — code pending
+
+
+
+ + +
+

+ 参考资源 + References +

+

+ 进一步学习和追踪 + Further reading and tracking +

+ +
+
+

Awesome 3D Scene Generation

+

959 Stars，最全面的论文+代码列表959 Stars, the most comprehensive paper + code list

+ GitHub → +
+
+

WorldMesh Paper

+

arXiv 2603.22972，等代码释出后第一时间复现arXiv 2603.22972 — watch for code release

+ arXiv → +
+
+

WorldMesh GitHub

+

Watch 仓库，代码发布后会有通知Watch this repo for code release notification

+ GitHub → +
+
+
+ +
+

+ 3D 场景生成 · 开源复现计划 · 源码保存于 repos/ + 3D Scene Generation · Open-Source Reproduction Plan · Source saved in repos/ +

+

+ 本地服务Local server: python3 -m http.server 4330 +

+
+ + +