Skip to content

Latest commit

 

History

History
49 lines (39 loc) · 2.87 KB

README.md

File metadata and controls

49 lines (39 loc) · 2.87 KB
Raw

High-quality Spectral Image Resonstruction and Segmentation Dataset (HSIRS)

[Paper]

Overview

We introduce HSIRS, a large scale dataset of hyper-spectral images along with corresponding manually annotated segmentation maps for material characterization and classification based on spectral signature. Such data can be used to simulate any type of spectrometer and to train DNNs end-to-end for spectral reconstruction and image segmentation tasks. HSIRS features scenes containing real and fake (made of polyster, plastic or ceramic) food items with different backgrounds and scene layouts, some scnes contain also colorcheckers. Spectral bands are sequentially captured using a VariSpecTM tunable color filter and the scene is illuminated with 4 Halogen light sources.

alt text Figure 1: HSIRS can be used as is or to emulate different snapshot spectrometers.

HSIRS Stats

Nbr. of images Spatial res. Spectral range (nm) Spectral res. / Nbr. bands Nbr. semantic classes Illumination Image sensor
592 2048x2048 470 - 700 7 nm / 33 bands 40 4 Hologen lamps Sony IMX287

alt text Figure 2: (Left) Number of fake/real occurences of each food item class. (Right) Sample images in sRGB space with overlaid segmentaion maps.

Dataset Structure

In the main HSIRS/ folder there are 592 sub-folders for each captured scene. Each with a name scene_name and within every sub-folder there are 35 files:

  • scene_name_xxx_nm.png: 8-bit gray scale image for a given spectral band (total of 33 bands).
  • scene_name_rgb.png: The sRGB image obtained from the hyper-spectral cube using the CIE 1931 conversion norm.
  • scene_name_seg_map.png: 8-bit gray scale image of the segmentation map.

Each scene name starts with YYYYMMDDHHMMSS indicating the year,month,day, and time of capture.

The segmentation map for each scene contain classes' IDs [0,1,2,..,40], 0 being the ID for the back ground class.

In addition to that, labels.txt is a text file containing the names for all the semantic classes present in the dataset

Download

Download via SharePoint from here

Citation

If you use this dataset in your research please cite our work:

@ARTICLE{TCI_CTIS,
  author={Mel, Mazen and Gatto, Alexander and Zanuttigh, Pietro},
  journal={IEEE Transactions on Computational Imaging}, 
  title={Joint Reconstruction and Spatial Super-resolution of Hyper-Spectral CTIS Images via Multi-Scale Refinement}, 
  year={2024},
  volume={},
  number={},
  pages={1-14},
  doi={10.1109/TCI.2024.3446230}
}

License

See LICENSE.txt