COPA-SG
收藏资源简介:
COPA-SG是一个合成的场景图数据集,它具有高度精确的地面实况和所有对象之间详尽的关联注释。该数据集引入了参数化和原型关联两种新的基本概念,为场景图提供了更精细的表示。COPA-SG包含超过8600万个关联注释,这些关联注释是基于一套严格的确定性规则提取的,不依赖于主观的人类解释。COPA-SG可用于评估现有场景图模型,以检验它们是否适合下游使用,并通过引入参数化和原型关联,增强了场景图的推理和规划能力。
COPA-SG is a synthetic scene graph dataset with highly precise ground truth and exhaustive association annotations between all objects. This dataset introduces two novel fundamental concepts: parameterized and prototypical associations, which enable finer-grained representations for scene graphs. COPA-SG contains over 86 million association annotations, which are extracted based on a set of strict deterministic rules without relying on subjective human interpretation. COPA-SG can be used to evaluate existing scene graph models to verify their suitability for downstream applications, and enhances the reasoning and planning capabilities of scene graphs via the introduction of parameterized and prototypical associations.




