Core Insights - Huawei, in collaboration with Shanghai Jiao Tong University and Huazhong University of Science and Technology, has launched a world model called WordGrow, capable of generating large indoor scenes of up to 1800 square meters in size [1][12]. Group 1: Technology Overview - WordGrow can generate a 272 square meter indoor scene in just 30 minutes using a single A100 GPU, achieving a speed six times faster than similar technologies [11]. - The model employs three core technologies: precise data preprocessing, a 3D block completion mechanism, and a coarse-to-fine generation strategy, which collectively enhance the quality and coherence of generated scenes [9][10]. Group 2: Performance Metrics - Experimental data indicates that WordGrow achieves state-of-the-art (SOTA) geometric reconstruction metrics, with a low Fréchet Inception Distance (FID) of 7.52, significantly outperforming mainstream methods like SynCity and BlockFusion [12]. - Even when expanded to a 7x7 block ultra-large scene, the edge quality remains stable, demonstrating the robustness of the model [10]. Group 3: Team and Research Background - The primary authors of the research are Sikuang Li and Chen Yang from Shanghai Jiao Tong University, who completed the study during their internship at Huawei. The research focuses on computer vision and computer graphics [13].

闻乐 发自 凹非寺 量子位 | 公众号 QbitAI AI大house真来了。 华为联合上海交通大学、华中科技大学推出了世界模型 WordGrow ，可以生成 1800㎡ 超大室内场景 （19x39块） ，单卡30分钟就跑了 272㎡。 并且场景具备连贯的几何拓扑和照片级真实感外观，智能体的导航则是在复杂空间布局中自主规划路径。 里面的虚拟人还能顺畅导航，不带迷路的。 (小声说：大平层确实需要导航) 还有一些方法最多只能造单个房间，扩展成套房就卡壳； 这可以说是走到哪建到哪了，那场景是怎么搭的？ 具备连贯的几何拓扑和照片级真实感外观 以前想造个像样的3D大场景，坑不少。 比如有些技术先靠2D模型画张图，再硬掰成3D，结果换完视角一看，沙发腿歪了、墙壁纹理断了…… 更离谱的是没有布局逻辑——出现冰箱塞进卧室，床摆在厨房的情况。 现在，WorldGrow来搞装修了（bushi），用三个核心技术填坑。 第一步是先做 数据精准预处理 ，从3D-FRONT这类大规模数据集里提取优质样本，用Blender执行场景切片，通过布尔交集对场景进行区块 切分，再靠occupancy检测确保区块内容密度 （可见内容≥95%） 。 ...