KADID-10k: A Large-scale Artificially Distorted IQA Database

By Hanhe Lin, Vlad Hosu, Dietmar Saupe. International Workshop on Quality of Multimedia Experience, 2019

634 citations recorded by the source corpus.

Current artificially distorted image quality assessment (IQA) databases are small in size and limited in content. Larger IQA databases that are diverse in content could benefit the development of deep learning for IQA. We create two datasets, the Konstanz Artificially Distorted Image quality Database (KADID-10k) and the Konstanz Artificially Distorted Image quality Set (KADIS-700k). The former contains 81 pristine images, each degraded by 25 distortions in 5 levels. The latter has 140,000 pristine images, with 5 degraded versions each, where the distortions are chosen randomly. We conduct a subjective IQA crowdsourcing study on KADID-10k to yield 30 degradation category ratings (DCRs) per image. We believe that the annotated set KADID-10k, together with the unlabelled set KADIS-700k, can enable the full potential of deep learning based IQA methods by means of weakly-supervised learning.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/beffd42ea6c0adbf437afa989fd0f9924006101b

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Uncertainty-Aware Blind Image Quality Assessment in the Laboratory and Wild (2020, methodology): “NIQE and its improved version ILNIQE do not perform well on realistic distortions and challenging synthetic distortions in KADID-10K [6], despite the original goal of handling arbitrary distortions.”
Image Quality Assessment: Unifying Structure and Texture Similarity (2020, methodology): “We choose the large-scale IQA dataset KADID-10k [59] as the training set, which contains 81 reference images, each of which is distorted by 25 distortion types at 5 distortion levels.”
MetaIQA: Deep Meta-Learning for No-Reference Image Quality Assessment (2020, methodology): “To validate the generalization performance of our meta-model for unknown distortions, we compare our method with six state-of-the-art general-purpose NR-IQA methods by using the Leave-One-Distortion-Out cross validation on TID2013 [38] and KADID-10K [27] databases.”
MANIQA: Multi-dimension Attention Network for No-Reference Image Quality Assessment (2022, methodology): “To verify our method’s generalization, we have conducted experiments on the PIPAL and other four datasets (TID2013, LIVE, CSIQ, and KADID-10K).”
DreamSim: Learning New Dimensions of Human Visual Similarity using Synthetic Data (2023, methodology): “We evaluate various metrics on the TID2013 [64] and KADID-10k [49] image quality assessment (IQA) benchmarks and report results in [64] and KADID-10k [49] While LPIPS and FSIM are best-performing overall, DreamSim is still competitive in IQA despite not having been tuned on low-level distortions.”
TOPIQ: A Top-Down Approach From Semantics to Distortions for Image Quality Assessment (2023, methodology): “All model variants are trained on KADID-10k and tested on CSIQ and TID2013.”
KonIQ-10k: An Ecologically Valid Database for Deep Learning of Blind Image Quality Assessment (2019, background): “KADID-10k [26] was proposed to address the limitation with 10,125 distorted images.”
No-Reference Image Quality Assessment via Transformers, Relative Ranking, and Self-Consistency (2021, methodology): “For synthetically distorted datasets, we use LIVE [78], CSIQ [79], TID2013 [80], and KADID-10K [81], where among them KADID has the most number of distortion typesand distorted images.”