Falling Tower
收藏资源简介:
Falling Tower数据集是一个用于物理推理任务的稳定性检测QA数据集,包含模拟和真实世界的场景。该数据集通过模拟生成,涵盖了对象属性、位置和动态等信息,并生成了相关的问答对。数据集的创建旨在通过模拟数据增强视觉语言模型的物理推理能力,特别是在动态环境中的对象行为预测和稳定性分析。该数据集的应用领域主要集中在物理推理和视觉语言模型的性能提升上,旨在解决当前模型在物理和因果推理方面的不足。
The Falling Tower Dataset is a question-answering (QA) dataset for stability detection tasks in physical reasoning, encompassing both simulated and real-world scenarios. It is generated through simulation, incorporates information including object attributes, positions, and dynamics, and produces corresponding question-answer pairs. The core goal of developing this dataset is to augment the physical reasoning abilities of visual language models (VLMs) with simulated data, specifically focusing on object behavior prediction and stability analysis in dynamic environments. Its application domains primarily center on physical reasoning and enhancing the performance of visual language models, aiming to address the current shortcomings of existing models in physical and causal reasoning.

- 1Synthetic Vision: Training Vision-Language Models to Understand Physics多伦多大学 · 2024年



