CausalVerse
收藏资源简介:
CausalVerse是一个用于因果表示学习的高保真模拟视觉数据集,结合了真实的视觉复杂性和对地面真实因果生成过程的完整访问。数据集包括约20万张图像和3亿个视频帧,涵盖静态图像生成、动态物理模拟、机器人操作和交通情况分析四个领域。CausalVerse为因果表示学习提供了一个全面且实用的测试平台,旨在解决现实世界中的问题,例如理解高维非结构化数据的因果因素。
CausalVerse is a high-fidelity simulated visual dataset for causal representation learning, which combines real-world visual complexity with full access to the ground-truth causal generation process. The dataset includes approximately 200,000 images and 300 million video frames, covering four domains: static image generation, dynamic physical simulation, robotic manipulation, and traffic scenario analysis. CausalVerse provides a comprehensive and practical testbed for causal representation learning, aiming to solve real-world problems such as understanding causal factors in high-dimensional unstructured data.

- 1通过卡内基梅隆大学, Mohamed bin Zayed 人工智能大学 · 2025年



