Three datasets of visual quality assessments (ACR) with subject and country annotations
We have investigated the potential effects of national differences on image and video quality assessment using discrete rating scales. Drawing on cultural psychology, we hypothesized that observers from different countries may exhibit distinct response styles in interpreting and applying the five-level absolute and degradation category rating scales (ACR, DCR). For our study, we adapted state-of-the-art statistical models for two of our earlier large-scale image quality datasets (KonIQ-10k, KADID-10k) and one based on the Netflix International Video Dataset (NIVD).
On this page we make available these curated datasets. Besides ACR ratings they contain anonymized subject and country annotations required to analyze country-specific rating behavior and to apply subject models like that of ITU P.914 resp. BT.500.
The original NIVD dataset is based on interval scale ratings using the SAMVIQ protocol. As shown in our paper, it is reasonable to quantize these continuous ratings to match the discrete ACR scale. These quantized ratings are contained in the dataset here. For the original raw data, please consult the Netflix dataset repository.
Cite us
VQA@Country is freely available to the research community. If you use our database (and NIVD) in your research, please cite it as follows:
@inproceedings{saupe2024national,
title={National differences in image quality assessment: An investigation on three large-scale IQA datasets},
author={Saupe, Dietmar and Del Pin, Simon Hviid},
booktitle={16th International Conference on Quality of Multimedia Experience (QoMEX)},
pages={},
year={2024},
organization={IEEE}
}
For the original NIVD dataset see Christos G Bampis, Lukas Krasula, Zhi Li, and Omair Akhtar, “Measuring and predicting perceptions of video quality across screen size with crowdsourcing,” in 15th International Conference on Quality of Multimedia Experience (QoMEX). IEEE, 2023, pp. 13–18.
The ACR.CSV files contain 4 columns giving subject, country, image, and rating
The countries are indicated by integers 1,2,... The INFO.CSV files show for each country code the name of the country and a summary of the number of subjects and ratings from each country.
NIVD.ACR.CSV
This file contains the NIVD interval scale ratings, quantized to the ACR scale. Subjects are from four countries which are explicitly given for each rating.