Event-based Depth Completion Datasets

CVRP Lab, National University of Singapore

Overview

The TOFDC dataset is collected by the time-of-flight (TOF) sensor and RGB camera of a Huawei P30 Pro. The ground truth depth maps are captured by the Helios TOF camera. This dataset covers various scenes such as texture, flower, body, and toy, under different lighting conditions and in open space. It has 10k 512×384 RGB-D pairs for training and 560 pairs for evaluation.


Fig.1 Distribution of different scenarios in our TOFDC.

Split

Train: 10,000 (RGB + sparse depth + GT depth)
Test: 560       (RGB + sparse depth + GT depth)
Data folder:
-test
        -rgb
        -tgv_gt
        -tof
-train
        -1_texture
        -2_flower
        -3_openspace
        -4_light
        -5_Video

Experiments


Fig.3 Visual results of RigNet and TPVD on TOFDC.

Copyright

The TOFDC dataset is exclusively available for academic purposes. Any researcher utilizing the TOFDC dataset is required to adhere to the following license:

1. This dataset is strictly for non-commercial use.
2. The entirety of the TOFDC dataset is copyrighted by the PCA Lab at Nanjing University of Science and Technology. This stipulates that unsers must credit the work in the manner delineated by the authors.
3. If users modify, adapt, or build upon this work, they may distribute the resulting work solely under an identical license.

Download


To facilitate research, we offer direct access to the data download source, eliminating the need for applications. However, we kindly request that users adhere to the license stipulated above.

Camera intrinsics (3648*2736): fx=2828.5, cx=1816, fy=2837.9, cy=1394.9

Camera intrinsics (512*384): fx=396.9825, cx=254.8772, fy=398.3018, cy=195.7754

Note that the depth maps of TOFDC are measured in millimeters (mm).

Baidu Disk (nust)
Google Drive

TOFDSR for depth super-resolution:

Baidu Disk (a286)
Hugging Face

In addition, if you use our dataset, please cite:

Citation

                    
@article{yan2025event,
  title={Event-Driven Dynamic Scene Depth Completion},
  author={Yan, Zhiqiang and Jiao, Jianhao and Wang, Zhengxue and Lee, Gim Hee},
  journal={arXiv preprint arXiv:2505.13279},
  year={2025}
}

Contact

For any questions, please contact yanzq@nus.edu.sg

© Zhiqiang Yan | Last updated: Nov. 2025