TaRF
收藏资源简介:
TaRF 是由密歇根大学、耶鲁大学和加州大学伯克利分校联合创建的视触融合场景数据集,旨在将视觉与触觉信号对齐至共享的三维空间。该数据集包含 19.3k 对齐的视觉与触觉样本,覆盖 13 个普通场景,如办公室、走廊和户外环境。数据采集通过结合神经辐射场(NeRF)和触觉传感器完成,利用多视图几何方法校准视觉与触觉信号,实现空间对齐。TaRF 的创建过程包括场景的多视角视觉重建和同步采集触觉信号,最终通过扩散模型生成未直接采样的触觉信号。该数据集可用于触觉信号估计、触觉定位和材料属性理解等任务,为机器人交互和虚拟世界构建提供重要支持。
TaRF is a visuo-tactile fusion scene dataset jointly created by the University of Michigan, Yale University, and the University of California, Berkeley. It aims to align visual and tactile signals into a shared three-dimensional space. This dataset contains 19.3k aligned visual and tactile samples, covering 13 common scenarios such as offices, corridors, and outdoor environments. Data collection is completed by combining Neural Radiance Field (NeRF) and tactile sensors, using multi-view geometry methods to calibrate visual and tactile signals to achieve spatial alignment. The creation process of TaRF includes multi-view visual reconstruction of the scene and synchronous collection of tactile signals, and finally generates tactile signals that were not directly sampled via diffusion models. This dataset can be used for tasks such as tactile signal estimation, tactile localization, and material property understanding, providing important support for robot interaction and virtual world construction.
Tactile-Augmented Radiance Fields (TaRF)
数据集信息
- 数据集名称:Tactile-Augmented Radiance Fields (TaRF)
- 作者:Yiming Dou, Fengyu Yang, Yi Liu, Antonio Loquercio, Andrew Owens
- 相关论文:Tactile-augmented radiance fields
数据集下载
使用说明
安装
- 安装Nerfstudio模块
- 安装Diffusion模块
数据准备与模型加载
- 下载COLMAP数据库
- 下载预训练的NeRF和Diffusion模型
运行与交互
- 启动Nerfstudio查看器
- 与TaRF交互
触觉信号估计
- 修改
bg_path参数 - 启动实时触觉估计器
训练自己的TaRF
- 训练触觉估计器
引用
@inproceedings{dou2024tactile, title={Tactile-augmented radiance fields}, author={Dou, Yiming and Yang, Fengyu and Liu, Yi and Loquercio, Antonio and Owens, Andrew}, booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}, pages={26529--26539}, year={2024} }




