GenerativeWorld Renderer Dataset
收藏资源简介:
该数据集由Alaya Studio等机构联合创建,是一个基于AAA游戏引擎构建的大规模动态渲染数据集,包含400万帧720p/30fps的连续视频序列,同步提供RGB图像和五种G-buffer通道(深度、法线、反照率、金属度和粗糙度)。数据通过创新的双屏拼接捕获技术获取,涵盖多样化的场景、天气条件和运动模糊变体,具有丰富的时空连贯性。该数据集专为双向渲染任务设计,既支持复杂场景下的几何与材质分解研究,也能促进基于G-buffer引导的高保真视频生成。其应用领域包括计算机视觉中的逆向渲染、可控视频合成以及游戏内容生成等前沿方向。
This dataset was jointly created by Alaya Studio and other affiliated institutions. It is a large-scale dynamic rendering dataset built using AAA game engines, containing 4 million frames of continuous video sequences at 720p/30fps, with synchronized RGB images and five types of G-buffer channels: depth, normal, albedo, metallic, and roughness. The dataset was collected through an innovative dual-screen stitching capture technique, covering diverse scenes, weather conditions, and motion blur variants, and exhibits rich spatiotemporal coherence. Specifically designed for bidirectional rendering tasks, this dataset supports research on geometry and material decomposition in complex scenes, while also facilitating G-buffer-guided high-fidelity video generation. Its application domains cover cutting-edge research directions including inverse rendering in computer vision, controllable video synthesis, and game content generation.

- 1Generative World RendererAlaya Studio; 盛大AI研究院东京分部; 国立台湾大学; 东京大学; 国立阳明交通大学 · 2026年



