shuinb/SHU_SinglePhoton_Data

The Single-Photon Real and Simulated Dataset

Overview

We present the Single-Photon Real and Simulated Dataset, a comprehensive dataset designed for single-photon imaging research. It encompasses two distinct data components: real-world single-photon data and simulated single-photon data. The dataset comprises 11 object categories, which cover common urban and natural scene elements, including Building, Bike, Bin, Roadblock, Human, Ship, Firehydrant, Apple, Banana, Cap, and Waterbottle.

Single-Photon Real Data

Each category in the real data subset includes 400 npz files, providing rich depth information for computer vision applications. The dataset features diverse real-world scenarios with variations in lighting conditions, object scales, and viewpoints. Every npz file contains two key-value pairs:

depth_ssp: Single-photon estimated depth maps derived from time-of-flight measurements

Z_true: Ground-truth real depth maps with sub-millimeter precision

Single-Photon Simulated Data

The single-photon simulated data is obtained by simulating RGBD data through a single-photon simulation algorithm. Captured by a RealSense D435i device, the resolution of the simulated data is 128 * 72, downsampled from 640 * 360. This simulated data also consists of 11 categories, with a total of 106,458 npz files. The number of files for each category is detailed as follows:

Apple: 12392 files

Apple: 11302 files

Bin: 8852 files

Bin: 8688 files

Cap: 10204 files

Bike: 6024 files

Firehydrant: 10304 files

Human: 8688 files

Roadblock: 4548 files

Ship: 7256 files

Waterbottle: 18189 files

Citation

@misc{zhang2025labelefficient,

title={Label-efficient Single Photon Images Classification via Active Learning},

author={Zili Zhang and Ziting Wen and Yiheng Qiang and Hongzhou Dong and Wenle Dong and Xinyang Li and Xiaofan Wang and Xiaoqiang Ren},

year={2025},

eprint={2505.04376},

archivePrefix={arXiv},

primaryClass={cs.CV}

}

Direct HTTP Download(on HuggingFace)

Download (373 MB, real data .npz files；2.91 GB, Simulated data .npz data)

Category List

The dataset includes the following 11 object categories, which are carefully selected to represent a wide range of scenarios in single-photon imaging research:

Building

Bike

Bin

Roadblock

Human

Ship

Firehydrant

Apple

Banana

Cap

Waterbottle

Each category contains npz files with consistent naming conventions (e.g., building_0001.npz, bike_0350.npz) for easy data management. In the real data subset, each category has 400 files, while the simulated data subset has varying numbers of files per category as detailed above.

Dataset Statistics

Attribute	Real Data Value	Simulated Data Value
Number of Categories	11	11
Files per Category	400 npz files	Varying (see above details)
Total Files	4,400 npz files	106,458 npz files
File Format	NPZ (compressed numpy archive)	NPZ (compressed numpy archive)
Key-value Pairs	depth_ssp (estimated depth),Z_true (ground-truth depth)	depth_ssp (estimated depth),Z_true (ground-truth depth)
Depth Map Resolution	55x55 pixels	128x72

The dataset maintains a balanced distribution across all categories in the real data subset, with approximately equal representation of small (e.g., apple, banana), medium (e.g., bike, cap), and large (e.g., building, ship) objects. In the simulated data subset, while the number of files per category varies, it still provides a comprehensive coverage of the 11 categories for research purposes.

License

The Single-Photon Real and Simulated Dataset is released under a Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0). This allows use, sharing, adaptation, and distribution for both commercial and research purposes, provided that:

Appropriate credit is given to the original dataset authors.

Any modified versions are distributed under the same license.

Copyright for the raw sensor data remains with the contributing institutions. Users are responsible for adhering to privacy and ethical guidelines when using human-related data (Category 5: Human).