This is an extended VirtualHome based on VirtualHome v2.2 with some additional functions.
For more information about original VirtualHome v2.2, please see here.
- We added actions that can be executed in the Unity simulator. That is, many motions not supported by the original VirtualHome can now be executed.
- Several new cameras and three new camera modes have been added.
- It is now possible to output JSON data in frame-by-frame.
- The 2D bounding boxes of objects can now be output.
- A convenient jupyter notebook is provided to run a large number of simulations based on the script data and save the results.
Download the VirtualHome UnitySimulator executable and move it under simulation/unity_simulator.
To test the simulator in a local machine, double click the executable.
To use the functions added by VirtualHome-AIST, please see here.
A framework to convert VirtualHome execution results into knowledge graphs.
For more information, please visit here.
Datasets created using VirtualHome-AIST and VirtualHome2KG are available here.
A visualization tool was developed to simultaneously check the generated videos and corresponding knowledge graphs.
Please visit here.
Shusaku Egami, Takanori Ugai, Swe Nwe Nwe Htun, Ken Fukuda: "VHAKG: A Multi-modal Knowledge Graph Based on Synchronized Multi-view Videos of Daily Activities." Proceedings of the 33rd ACM International Conference on Information and Knowledge Management (CIKM2024), to appear, 2024.10 [preprint]
@InProceedings{egami24vhakg,
author="Shusaku Egami
and Takanori Ugai
and Swe Nwe Nwe Htun
and Ken Fukuda",
title="{VHAKG}: A Multi-modal Knowledge Graph Based on Synchronized Multi-view Videos of Daily Activities",
booktitle="Proceedings of the 33rd ACM International Conference on Information and Knowledge Management",
note="to appear",
year="2024",
doi="10.1145/3627673.3679175"
}