A carefully crafted cross-resolution image quality database

The perceptual quality of images changes with the presentation scale. Down-sampling attenuates impairments, e.g., blurs or compression artifacts, which can positively affect the impression evoked in subjective studies. To accurately predict perceptual image quality, IQA methods must account for the perceptual label shifts in the ground truth.

KonX is the first database that disentangles perceptual quality and annotation resolution.

​Cite us

KonX is freely available to the research community. If you use our database in your research, you can cite it as follows:

@article{konx,
  title={Konx: cross-resolution image quality assessment},
  author={Wiedemann, Oliver and Hosu, Vlad and Su, Shaolin and Saupe, Dietmar},
  journal={Quality and User Experience},
  volume={8},
  number={1},
  pages={8},
  year={2023},
  publisher={Springer}
}

Downloads

KonX Dataset (.zip)

752 MB

The scores files

We include several .csv files with splits (views) of the data for your convenience. The complete information is included in `scores_aggregated_sets.csv`:

• image_name: the image file name, the prefix indicates:
  • the source (Koniq10k or Pixabay)
  • the resolution
• MOS: Mean Opinion Scores of the subjective study
• ​std: ​Standard Deviation of the MOS
• set: image resolution (redundant information for convenience)

Please read our paper for more details on how the data was generated, screened for reliability, etc.