Subjective and Objective Quality Assessments of AI-generated Images for Language E-Learning

This repository hosts a dataset of AI-generated images for Language E-Learning. The dataset consists of 1,000 generated images, created using five text-to-image models: four from Stability AI (SD 1.5, SDXL 1.0, SDXL Turbo, and SD 2.1), and DALL·E 3 from OpenAI. Each generator contributed 200 images, organized as follows:Image index no: 1 - 200 (SD 1.5)image index no: 201 - 400 (SDXL 1.0)image index no: 401 - 600 (SDXL Turbo)image index no: 601 - 800 (SD 2.1)image index no: 801 - 1000 (DALL.E 3)The images in this dataset represent four distinct styles (anime, abstract, realistic and cartoon styles), curated to support language learning. Each image was generated using a complete English sentence as the prompt, making them suitable for developing the four fundamental language learning skills. A subjective experiment was conducted on the dataset, and Mean Opinion Scores (MOS) were estimated after post-processing the human ratings in accordance with the ITU-R BT.500-15 guidelines. This dataset is now available and can be used to support e-learning.Please cite this paper if you use the dataset.