VNIRSR Hyperspectral Reflectance Dataset

VNIRSR is a visible–near-infrared (VNIR) hyperspectral reflectance image dataset introduced in
Two-stage framework for reconstructing spectral images from diffraction-blurred images
[paper].

The dataset contains hyperspectral reflectance images and their corresponding RGB images.

• Hyperspectral reflectance data are provided in the mat/ directory.
• RGB images are provided in the RGB/ directory. They are simulated from the hyperspectral data using a D65 standard illuminant and the CIE 1931 color matching functions.

---

Dataset Characteristics

• Spatial resolution: 936 × 960
• Spectral range: 400 nm – 900 nm
• Spectral interval: 10 nm
• Number of spectral channels: 51
• Number of scenes:
  • 217 artificially constructed scenes
  • 113 outdoor scenes
  • Total: 330 hyperspectral data cubes

---

Citation

If you use this dataset in your research, please cite the following paper:

@article{xu2025two,
  title   = {Two-stage framework for reconstructing spectral images from diffraction-blurred images},
  author  = {Xu, Hao and Zhuge, Zhengyue and Chen, Bingkun and Xu, Nan and Xu, Jiahui and Zhang, Gongfu and He, Jiajian and Xu, Zhihai and Li, Qi and Chen, Yueting},
  journal = {Optics and Lasers in Engineering},
  volume  = {186},
  pages   = {108789},
  year    = {2025},
  publisher = {Elsevier}
}