CAUSAL3D
收藏资源简介:
CAUSAL3D是一个针对视觉数据因果学习的新型综合基准,由Case Western Reserve University设计。该数据集整合了结构化数据(表格)和相应的视觉表示(图像),包含19个3D场景数据集,捕捉多样的因果关联、视角和背景,支持不同复杂场景下的评估。数据集由物理一致的场景和假设场景组成,物理场景基于现实世界的物理规则,假设场景则创造性地设计了因果关联。每个场景包含2至5个变量,共有10K个样本,旨在推动计算机视觉和因果推理领域的发展。
CAUSAL3D is a novel comprehensive benchmark for causal learning on visual data, designed by Case Western Reserve University. This dataset integrates structured data (in tabular format) and their corresponding visual representations (images), comprising 19 3D scene datasets that capture diverse causal associations, viewing angles and backgrounds, enabling evaluation across various complex scenarios. The dataset consists of physically consistent scenes and hypothetical scenarios: physical scenes are grounded in real-world physical laws, while hypothetical scenarios creatively construct causal associations. Each scene contains 2 to 5 variables, with a total of 10,000 samples. This benchmark aims to advance the development of the fields of computer vision and causal reasoning.

- 1CAUSAL3D: A Comprehensive Benchmark for Causal Learning from Visual DataCase Western Reserve University · 2025年



