| license: other | |
| license_name: mixed-upstream-licenses | |
| license_link: LICENSE | |
| task_categories: | |
| - robotics | |
| # 3DWay Dataset | |
| This repository contains the training data for **3DWay: Generalizing Robot Manipulation via 3D Consistent Waypoints**. | |
| **Paper:** [3DWay: Generalizing Robot Manipulation via 3D Consistent Waypoints](https://huggingface.co/papers/2609.08224) | |
| **Code:** [github.com/ziqin-h/3DWay](https://github.com/ziqin-h/3DWay) | |
| ## Contents | |
| The dataset is built from three robotics datasets: | |
| ```text | |
| droid.json | |
| droid.tar | |
| rh20t.json | |
| rh20t.tar | |
| rlbench.json | |
| rlbench.tar | |
| ``` | |
| Each `.json` file contains the corresponding training annotations, while each `.tar` archive contains the associated image data. | |
| ## Extraction | |
| The image archives can be extracted directly with `tar`: | |
| ```bash | |
| tar -xvf droid.tar | |
| tar -xvf rh20t.tar | |
| tar -xvf rlbench.tar | |
| ``` | |
| ## Project Page | |
| https://ziqin-h.github.io/3DWay/ | |
| ## Citation | |
| ```bibtex | |
| @inproceedings{3dway2026, | |
| title = {3DWay: Generalizing Robot Manipulation via 3D Consistent Waypoints}, | |
| author = {Huang, Ziqin and Li, Yingyue and Zhang, Chenyangguang and Zhang, Ruida and Chen, Yuxin and Wang, Gu and Liu, Xingyu and Tomizuka, Masayoshi and Ji, Xiangyang}, | |
| booktitle = {European Conference on Computer Vision (ECCV)}, | |
| year = {2026} | |
| } | |
| ``` | |
| ## License | |
| This repository contains data derived from multiple upstream datasets and is released under **mixed upstream licenses**. Please refer to the `LICENSE` file for details. |