prajnan/preprocessed_2d3d

2D3D-MATR Preprocessed Datasets (7-Scenes & RGB-D Scenes v2)

This repository hosts the preprocessed image-to-point-cloud data used by 2D3D-MATR (ICCV 2023) for training and evaluation of 2D–3D registration between RGB images and point clouds. The data is derived from two upstream datasets:

• Microsoft 7-Scenes
• RGB-D Scenes v2 (University of Washington)

Frames have been sampled and paired with corresponding point cloud fragments, with ground-truth 2D–3D correspondences and relative poses, following the protocol described in the 2D3D-MATR paper.

⚠️ This repository contains preprocessed derivatives of the original datasets. The licensing terms of the original data still apply. Please read the License section carefully before downloading.

Dataset Structure

data/Datasets/
└── 7Scenes/
|   ├── metadata/
|   └── data/
|       ├── chess/
|       ├── fire/
|       ├── heads/
|       ├── office/
|       ├── pumpkin/
|       ├── redkitchen/
|       └── stairs/
└── RGBDScenesV2/
    ├── metadata/
    └── data/
        ├── rgbd-scenes-v2-scene_01/
        ├── ...
        └── rgbd-scenes-v2-scene_14/

Intended Use

These splits are intended for non-commercial academic research on image-to-point-cloud registration, cross-modal feature learning, and related 3D vision tasks. The preprocessed data is provided for reproducibility of the 2D3D-MATR's evaluation protocol.

License

This repository does not relicense the upstream data. Two distinct licenses apply, and users must comply with both:

7-Scenes (Microsoft Research)

The 7-Scenes data is distributed under the Microsoft Research License Agreement (MSR-LA) and is provided for non-commercial use only. By downloading any portion of the 7-Scenes-derived data in this repository, you accept the terms of the MSR-LA. The original license can be obtained from the 7-Scenes project page.

RGB-D Scenes v2 (University of Washington)

The RGB-D Scenes v2 dataset is publicly released by the University of Washington (Lai, Bo, and Fox) on the RGB-D Object Dataset website. The original release page does not specify an explicit license; it is widely used in the research community for academic, non-commercial purposes with attribution. Users wishing to redistribute or use the data outside of academic research should contact the original authors for explicit permission.

Preprocessing Scripts and Metadata

Any preprocessing scripts, metadata files, and split definitions authored as part of 2D3D-MATR and included in this repository are made available under the same terms as the original 2D3D-MATR release for research reproducibility. They are derivative works of the upstream datasets and inherit the upstream restrictions for the data portions.

If you intend to use this data for any commercial purpose, you must obtain permission from the respective dataset owners (Microsoft Research for 7-Scenes; the University of Washington for RGB-D Scenes v2).

Citation

If you use this preprocessed data, please cite the 2D3D-MATR paper:

@inproceedings{li20232d3d,
  title={2D3D-MATR: 2D-3D Matching Transformer for Detection-free Registration between Images and Point Clouds},
  author={Li, Minhao and Qin, Zheng and Gao, Zhirui and Yi, Renjiao and Zhu, Chenyang and Guo, Yulan and Xu, Kai},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={14128--14138},
  year={2023}
}

Please also cite the original dataset papers:

7Scenes

@inproceedings{shotton2013scene,
  title={Scene coordinate regression forests for camera relocalization in RGB-D images},
  author={Shotton, Jamie and Glocker, Ben and Zach, Christopher and Izadi, Shahram and Criminisi, Antonio and Fitzgibbon, Andrew},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  pages={2930--2937},
  year={2013}
}

RGBDScenesV2

@inproceedings{lai2014unsupervised,
  title={Unsupervised feature learning for 3D scene labeling},
  author={Lai, Kevin and Bo, Liefeng and Fox, Dieter},
  booktitle={2014 IEEE International Conference on Robotics and Automation (ICRA)},
  pages={3050--3057},
  year={2014},
  organization={IEEE}
}

Acknowledgements

We thank the authors of 7-Scenes (Microsoft Research) and RGB-D Scenes v2 (University of Washington) for releasing the original data, and the authors of 2D3D-MATR for the benchmark protocol, and the preprocessing data.