Back to Publications

Author(s) Hosu, V., Lin, H., Sziranyi, T., Saupe, D.
Title KonIQ-10k: An ecologically valid database for deep learning of blind image quality assessment
Abstract Deep learning methods for image quality assessment (IQA)are limited due to the small size of existing datasets. Extensive datasetsrequire substantial resources both for generating publishable content,and annotating it accurately. We present a systematic and scalable ap-proach to create KonIQ-10k, the largest IQA dataset to date consisting of10,073 quality scored images. This is the first in-the-wild database aim-ing for ecological validity, with regard to the authenticity of distortions,the diversity of content, and quality-related indicators. Through the useof crowdsourcing, we obtained 1.2 million reliable quality ratings from1,459 crowd workers, paving the way for more general IQA models. Wepropose a novel, deep learning model (KonCept512), to show an excel-lent generalization beyond the test set (0.921SROCC), to the currentstate-of-the-art database LIVE-in-the-Wild (0.825SROCC). The modelderives its core performance from the InceptionResNet architecture,being trained at a higher resolution than previous models (512×384). Acorrelation analysis shows that KonCept512 performs similar to having9subjective scores for each test image