A Dataset and Quality Evaluation of AI-generated Images for Online Learning

AI-generated content is becoming an integral part of online education. Researchers have made efforts to examine the influence of such content on user satisfaction by creating AI-generated datasets and conducting subjective experiments. While several existing datasets cover various aspects of human life, we have developed the first AI-generated image (AIGI) dataset specifically for online learning. Also, we conducted a crowd-based subjective experiment to obtain human ratings for both perceptual and alignment assessments of the AIGIs. Using the new dataset as a benchmark, we carried out an evaluation of a large number of quality models. In contrast to previous studies, the results in this study showed that existing quality models are not effective yet in predicting quality of the images in online learning.