Affogato
收藏资源简介:
Affogato是一个大规模的数据集,包含150,000个实例,每个实例都有开放词汇文本描述和相应的3D可利用性热图。该数据集跨越了一系列多样的对象和交互。Affogato数据集由Affogato-Engine流水线自动生成,该流水线利用多视图渲染和最先进的视觉语言模型来创建具有不同对象类别的优质可利用性注释。Affogato数据集旨在解决智能体理解和与其环境交互的能力问题,特别是解决细粒度部分级定位、多个有效交互区域引起的歧义以及大规模数据集稀缺的问题。
Affogato is a large-scale dataset containing 150,000 instances, each paired with open-vocabulary textual descriptions and corresponding 3D affordance heatmaps. This dataset covers a diverse range of objects and interactions. The Affogato dataset is automatically generated via the Affogato-Engine pipeline, which leverages multi-view rendering and state-of-the-art vision-language models to create high-quality affordance annotations across different object categories. The Affogato dataset aims to address the challenges in enabling AI agents to understand and interact with their environments, specifically resolving issues including fine-grained part-level localization, ambiguity induced by multiple valid interaction regions, and the scarcity of large-scale datasets.

- 1Affogato: Learning Open-Vocabulary Affordance Grounding with Automated Data Generation at Scale浦项科技大学 (POSTECH) · 2025年



