RUOXUAN123/

MindPower

This dataset is designed to evaluate the Theory of Mind (ToM) capability of vision-language models (VLMs). It includes data collected from two simulators: ThreeDWorld and VirtualHome.

	Data Sources




	


	1. ThreeDWorld

In ThreeDWorld, all data are collected through manual keyboard control.Each sample in this simulator contains:

a video the corresponding ground-truth annotation

In VirtualHome, the data are… See the full description on the dataset page: https://huggingface.co/datasets/RUOXUAN123/MindPower.